| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/GUDDADAMALLAPURA/D_540_5493_644156.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"d_dod\": \"2012-01-19\", \"gender\": \"F\", \"age\": 85, \"d_deceased_nme_kn\": \"ಪಾರವ್ವ ಬಸಪ್ಪ ದೇಸಾಯಿ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": \"ಪ್ರಭು ಬಸಪ್ಪ ದೇಸಾಯಿ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644156", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/TIPALAPURA/D_540_5493_644129.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 2, \"d_dod\": \"1994-10-25\", \"gender\": \"F\", \"age\": 65, \"d_deceased_nme_kn\": \"ಕನ್ನವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಹೊನ್ನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": \"ಸುರೇಶ ಮಲ್ಲಪ್ಪ ಓಲೇಕಾರ\", \"info_details\": \"Thippalapur\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644129", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/DUMMIHALA/D_540_5493_644174.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 5, \"d_dod\": \"2012-04-13\", \"gender\": \"M\", \"age\": 90, \"d_deceased_nme_kn\": \"ಹುಚ್ಚಪ್ಪ\", \"d_father_kn\": \"ಬಸಪ್ಪ ಸೋದ್ರಿ\", \"d_h_w_name_kn\": null, \"mother_namekn\": \"ಬಸವ್ವ\", \"info_name_kn\": \"ಲಕ್ಷ್ಮಪ್ಪ ಹುಚ್ಚಪ್ಪ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644174", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644237.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 18, \"d_dod\": \"2012-10-05\", \"gender\": \"F\", \"age\": 58, \"d_deceased_nme_kn\": \"ಗೌರವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಗುಡ್ಡಪ್ಪ ಚಿಕ್ಕನಗೌಡ್ರ\", \"mother_namekn\": null, \"info_name_kn\": \"ಗುಡ್ಡಪ್ಪ ಫಕ್ಕೀರಪ್ಪ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644237", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009354.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 9, \"b_dob\": \"2004-02-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹ್ಮದ್ ಅಫ್ತಾಬ್\", \"b_father_kn\": \"ಇಂಬಿಯಾ ಅಹ್ಮದ್ ಮುಕಾಶಿ\", \"mother_namekn\": \"ಗುಲಾಬಾ\", \"info_name_kn\": \"ಪರಮೇಶ್ವರಿ ಎಸ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009354", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644245.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 10, \"d_dod\": \"1970-03-02\", \"gender\": \"M\", \"age\": 65, \"d_deceased_nme_kn\": \"ಕರಿಯಪ್ಪ ದಿಲ್ಲೆಪ್ಪ ಕಾಕೋಳ\", \"d_father_kn\": \"ದಿಲ್ಲೆಪ್ಪ ಕಾಕೋಳ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ದಿಲ್ಲೆಪ್ಪ ಹುಚ್ಚಪ್ಪ ಕಾಕೋಳ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644245", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644252.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 3, \"d_dod\": \"2012-02-12\", \"gender\": \"F\", \"age\": 50, \"d_deceased_nme_kn\": \"ಗೌರವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಹನುಮಂತಪ್ಪ ಬಿ ಎರೆಸಿಮಿ\", \"mother_namekn\": null, \"info_name_kn\": \"ಪುಟ್ಟಪ್ಪ ದಿಲ್ಲೆಪ್ಪ ಹಿರೇಗೌಡ್ರ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644252", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009367.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 22, \"b_dob\": \"2004-02-20\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಫಜೀಲಾ ಬಾನು\", \"b_father_kn\": \"ಫಜಲು ರೆಹಮಾನ್\", \"mother_namekn\": \"ಹಸೀನಾ ಬಾನು\", \"info_name_kn\": \"ಫಜಲು ರೆಹಮಾನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009367", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MALLAURU/B_540_5493_060779.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 37, \"b_dob\": \"1958-08-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮುಕಮಸಾಬ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 38, \"b_dob\": \"1958-08-07\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ರಾಮಪ್ಪ ಮುನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 39, \"b_dob\": \"1958-08-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನೀಲಕಂಠಪ್ಪ\", \"b_father_kn\": \"ರಾಜಪ್ಪ ರಾಮಣ್ಣ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 40, \"b_dob\": \"1958-08-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕುಸುಮಾಬಾಯಿ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ ವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 41, \"b_dob\": \"1958-08-09\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚನ್ನತುಬಿ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 42, \"b_dob\": \"1958-08-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಂಗಳವ್ವ\", \"b_father_kn\": \"ವೀರಪ್ಪ ಸೋಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 43, \"b_dob\": \"1958-08-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮದರಬಿ\", \"b_father_kn\": \"ಜಮಾಲಾಬಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 44, \"b_dob\": \"1958-03-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮದನ್ಬಿ\", \"b_father_kn\": \"ಜಮಾಲಾಬಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 45, \"b_dob\": \"1958-08-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಶಿವಬಸಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 46, \"b_dob\": \"1958-08-18\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಾಗವ್ವ\", \"b_father_kn\": \"ವೀರಭದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 47, \"b_dob\": \"1958-08-20\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮುತ್ತಯ್ಯಸ್ವಾಮಿ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 48, \"b_dob\": \"1958-07-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಾಗೇಂದ್ರಪ್ಪ\", \"b_father_kn\": \"ಗುರುಪಾದಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 49, \"b_dob\": \"1958-08-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರೇಣುಕವ್ವ\", \"b_father_kn\": \"ನಾಗಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 50, \"b_dob\": \"1958-08-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಾವಿತ್ರವ್ವ\", \"b_father_kn\": \"ಶಂಕರಪ್ಪ ವೀರಭದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 51, \"b_dob\": \"1958-08-23\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಂಕ್ರವ್ವ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 52, \"b_dob\": \"1958-08-28\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪಿತವ್ವ\", \"b_father_kn\": \"ರಾಮಪ್ಪ ಕೃಷ್ಣಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 53, \"b_dob\": \"1958-03-31\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶೆಲಾವತಿ\", \"b_father_kn\": \"ವೀರನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 54, \"b_dob\": \"1958-09-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟವ್ವ\", \"b_father_kn\": \"ಸೋಮಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 55, \"b_dob\": \"1958-09-03\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚನ್ನಬಸವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 56, \"b_dob\": \"1958-09-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗುಡಪ್ಪ\", \"b_father_kn\": \"ನಿರಗಪ್ಪ ವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 57, \"b_dob\": \"1958-09-11\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದಿಲಾಶದ್ರ\", \"b_father_kn\": \"ಭುದುನಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 58, \"b_dob\": \"1958-09-12\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಾಗರತ್ನ\", \"b_father_kn\": \"ಮೂರ್ತಾಂಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 59, \"b_dob\": \"1958-09-13\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹೇಮಯ್ಯ\", \"b_father_kn\": \"ಶಿವಮೂರ್ತಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 60, \"b_dob\": \"1958-09-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪ್ರೇಮವ್ವ\", \"b_father_kn\": \"ಗಸುಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 61, \"b_dob\": \"1958-09-24\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಾರಂಬಿ\", \"b_father_kn\": \"ಗಸುಸಾಬ ಮೋದಿನಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 62, \"b_dob\": \"1958-01-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನರಸಪ್ಪ\", \"b_father_kn\": \"ಗಣೇಶಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 63, \"b_dob\": \"1958-09-24\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಇಸ್ಲಾಮಸಾಬ\", \"b_father_kn\": \"ಅಲ್ಲಾಭಾಷಾ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 27, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_060779", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644254.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"d_dod\": \"1986-03-20\", \"gender\": \"F\", \"age\": 65, \"d_deceased_nme_kn\": \"ಹನುಮವ್ವ\", \"d_father_kn\": \"ಪುಟ್ಟಪ್ಪ ಹಿರೇಗೌಡ್ರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಪುಟ್ಟಪ್ಪ ದಿಲ್ಲೆಪ್ಪ ಹಿರೇಗೌಡ್ರ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644254", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009376.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 31, \"b_dob\": \"2004-02-27\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸಂಗಮೇಶ\", \"b_father_kn\": \"ರಮೇಶ ಈಳಿಗೇರ\", \"mother_namekn\": \"ಮಂಜುಳಾ\", \"info_name_kn\": \"ಸಿ ವಿ ಶರವಣ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009376", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047269.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 5, \"b_dob\": \"1958-02-18\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕರಬಸವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಚನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1958-02-25\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಶಿವಪ್ಪ ಬಸನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1958-02-27\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕರಬಸವ್ವ\", \"b_father_kn\": \"ಭೀಮಪ್ಪ ಗೋವಿಂದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 8, \"b_dob\": \"1958-02-28\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಪುಟ್ಟಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 9, \"b_dob\": \"1958-03-04\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚನಬಸವ್ವ\", \"b_father_kn\": \"ಅಡಿವಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 10, \"b_dob\": \"1958-03-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮುರಡಪ್ಪ\", \"b_father_kn\": \"ಸಂಕಪ್ಪ ಕರನಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 6, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047269", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048558.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 8, \"b_dob\": \"1970-04-16\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದುರಗವ್ವ\", \"b_father_kn\": \"ಕರಿಯಪ್ಪ ಬುಡಕನಳ್ಳಿ\", \"mother_namekn\": \"ಸಾವಕ್ಕ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 9, \"b_dob\": \"1970-04-30\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗುಳ್ಳಪ್ಪ\", \"b_father_kn\": \"ನಾಗಪ್ಪ ಪೂಜಾರ\", \"mother_namekn\": \"ಬಸವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048558", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009378.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 33, \"b_dob\": \"2004-03-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹ್ಮದ್ ಶಬ್ಬೀರ್ ಅಮ್ಮಿನ ಬಾವಿ\", \"b_father_kn\": \"ಮಹ್ಮದ್ ಶಾಫಿ ಅಮ್ಮಿನ ಬಾವಿ\", \"mother_namekn\": \"ಗುಲಜಾರ ಬಾನು\", \"info_name_kn\": \"ಶಹತಾಜ್ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009378", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047990.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 9, \"b_dob\": \"1947-01-01\", \"gender\": \"T\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ರುದ್ರಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 10, \"b_dob\": \"1947-01-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹಿರಿಯವ್ವ\", \"b_father_kn\": \"ಮುದಕಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 11, \"b_dob\": \"1947-01-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಭೀಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 12, \"b_dob\": \"1947-01-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಉಜ್ಜಪ್ಪ\", \"b_father_kn\": \"ಸಿದ್ದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1947\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047990", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048559.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 10, \"b_dob\": \"1970-05-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಾಹಾದೇವಪ್ಪ\", \"b_father_kn\": \"ಹನುಮಂತಪ್ಪ ಭರಮಪ್ಪ ಹಾವೇರಿ\", \"mother_namekn\": \"ನಿಂಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 11, \"b_dob\": \"1970-05-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶುಭಾಸು\", \"b_father_kn\": \"ಪಕ್ಕೀರಪ್ಪ ಚಿಕ್ಕಪ್ಪ ಬಲ್ಲಳ್ಳಿ\", \"mother_namekn\": \"ಪಕ್ಕೀರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 12, \"b_dob\": \"1970-05-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚಂದ್ರಶೇಕ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ಭರಮಪ್ಪ ಚೌಡಪ್ಪನವರ\", \"mother_namekn\": \"ಎಲ್ಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 13, \"b_dob\": \"1970-05-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚಂದ್ರಶೇಕ\", \"b_father_kn\": \"ಭರಮಪ್ಪ ಭರಮಪ್ಪ ಗುರಣನವರ\", \"mother_namekn\": \"ಪುಟ್ಟವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 14, \"b_dob\": \"1970-05-23\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಿವಾಜಿ\", \"b_father_kn\": \"ತಿಪ್ಪಣ್ಣ ದಾಸಪ್ಪ ಅಳ್ಳಿಕಟ್ಟಿ\", \"mother_namekn\": \"ಪುಟ್ಟವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 5, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048559", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009361.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 16, \"b_dob\": \"2004-02-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹ್ಮದ್ ಶಯೀದ್\", \"b_father_kn\": \"ಮಹಮ್ಮದ್ ಅಸ್ಲಂ\", \"mother_namekn\": \"ಸಲೀಮಾ\", \"info_name_kn\": \"ಗಾಯತ್ರಿ ಕೆ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009361", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009350.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 5, \"b_dob\": \"2004-02-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಬೀಬಿ ಆಯಿಷಾ\", \"b_father_kn\": \"ಶಮೀವುಲ್ಲಾ ಶಾಮನಾ ಭಾಯಿ\", \"mother_namekn\": \"ಯಾಸ್ಮೀನ್\", \"info_name_kn\": \"ಈರಪ್ಪ ತಳವಾರ\", \"info_details\": null}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009350", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009365.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 20, \"b_dob\": \"2004-02-14\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಪಟೇಲ ಸಾಬ\", \"mother_namekn\": \"ಶಹೀದಾ ಬಾನು\", \"info_name_kn\": \"ಶಹತಾಜ್ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009365", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/KENGONDA/B_540_5493_061669.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 12, \"b_dob\": \"1958-02-11\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸುಂದ್ರವ್ವ\", \"b_father_kn\": \"ನಾಗೇಂದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 13, \"b_dob\": \"1958-07-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಿವಾನಂದ\", \"b_father_kn\": \"ಈರಪ್ಪ ಮಾಳಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 14, \"b_dob\": \"1958-07-15\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದ್ಯಾಮವ್ವ\", \"b_father_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 15, \"b_dob\": \"1958-08-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಯಾಲಪ್ಪ\", \"b_father_kn\": \"ಹೊನಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 16, \"b_dob\": \"1958-08-18\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಾಗವ್ವ\", \"b_father_kn\": \"ಹೊನಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 17, \"b_dob\": \"1958-08-21\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದುರ್ಗವ್ವ\", \"b_father_kn\": \"ಹನುಮಂತಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 18, \"b_dob\": \"1958-08-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರಾಮನಗೌಡ\", \"b_father_kn\": \"ಹನುಮಂತಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_061669", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009355.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 10, \"b_dob\": \"2004-02-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ತಯಾಮಿಯಾ ಬಾನು\", \"b_father_kn\": \"ಅಬ್ದುಲ್ ರಬಾ ಅಮ್ದುನವರ\", \"mother_namekn\": \"ಅಮೀರಬಿ\", \"info_name_kn\": \"ಈರಪ್ಪ ತಳವಾರ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009355", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644247.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 8, \"d_dod\": \"2012-05-02\", \"gender\": \"F\", \"age\": 60, \"d_deceased_nme_kn\": \"ಸೋಮವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ನಾಗಪ್ಪ ತಮ್ಮಣ್ಣನವರ\", \"mother_namekn\": null, \"info_name_kn\": \"ಸೋಮಣ್ಣ ನಾಗಪ್ಪ ತಮ್ಮಣ್ಣನವರ್\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644247", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_0048545.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1969-01-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸೆಟ್ಟೆಪ್ಪ\", \"b_father_kn\": \"ಕರಬಸಪ್ಪ ಶೆಂಕ್ರಪ್ಪ ಕಬ್ಬಿಣದ\", \"mother_namekn\": \"ಬಸವಣ್ಣೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1969-01-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕರಬಸಪ್ಪ\", \"b_father_kn\": \"ಪಕ್ಕೀರಪ್ಪ ಹನುಮಂತಪ್ಪ ರಿತ್ತಿ\", \"mother_namekn\": \"ನೀಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0048545", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048557.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 4, \"b_dob\": \"1970-03-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ವೀರಣ್ಣ\", \"b_father_kn\": \"ಕಲಿವೀರಪ್ಪ ಹನುಮಣ್ಣ ಪೂಜಾರ\", \"mother_namekn\": \"ಸಾವಂತ್ರೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1970-03-05\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಲೀಮಾ\", \"b_father_kn\": \"ಅಬ್ದುಲ್ ಸಾಹೇಬ ಮವಲಾಸಾಬ ಸಾಹೇಬ ಬ್ಯಾಡಗಿ\", \"mother_namekn\": \"ಸಾಹೇಬಿ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1970-03-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹಾದೇವಪ್ಪ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಹೇಮಣ್ಣ ಪೂಜಾರ\", \"mother_namekn\": \"ಪಾರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1970-03-24\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಿವಪ್ಪ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಪಕ್ಕೀರಪ್ಪ ಗೋಣೆಪ್ಪನವರ\", \"mother_namekn\": \"ಯಮನವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048557", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_0047268.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-01-02\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಬಸವಣ್ಣಪ್ಪ ಶಿವಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-01-04\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ಶಿವಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 3, \"b_dob\": \"1958-01-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರುದ್ರಪ್ಪ\", \"b_father_kn\": \"ಕಲ್ಲಪ್ಪ ಚನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-01-09\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಶಿವನಗೌಡ ಮಲ್ಲನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0047268", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/TIPALAPURA/D_540_5493_644128.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"d_dod\": \"2012-07-03\", \"gender\": \"M\", \"age\": 85, \"d_deceased_nme_kn\": \"ಗುಡ್ಡಪಾಪ ಯಲ್ಲಪ್ಪ ಚವಟಿ\", \"d_father_kn\": \"ಯಲ್ಲಪ್ಪ ಚವಟಿ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ತಿರಕಪ್ಪ ಎನ್ ಮುಳಗುಂದ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644128", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048563.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 24, \"b_dob\": \"1970-09-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚಿಕ್ಕಪ್ಪ\", \"b_father_kn\": \"ಹೊನ್ನಪ್ಪ ನಿಂಗಪ್ಪ ಹಾವೇರಿ\", \"mother_namekn\": \"ಎಲ್ಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 25, \"b_dob\": \"1970-09-19\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕೊಟ್ರವ್ವ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ಬಸಪ್ಪ ಗುರಣ್ಣನವರ\", \"mother_namekn\": \"ಶಾಂತವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 26, \"b_dob\": \"1970-09-27\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನೀಲಪ್ಪ\", \"b_father_kn\": \"ತಿರಕಪ್ಪ ನೀಲಪ್ಪ ಕನವಳ್ಳಿ\", \"mother_namekn\": \"ಹೊನ್ನವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 3, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048563", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644236.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 19, \"d_dod\": \"2001-04-14\", \"gender\": \"F\", \"age\": 75, \"d_deceased_nme_kn\": \"ನೀಲವ್ವ ಊರ್ಫ್ ನೀಲಮ್ಮ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಗುಡ್ಡಪ್ಪ ಗೊಡ್ಡೆಮ್ಮೆ\", \"mother_namekn\": null, \"info_name_kn\": \"ಎಂ ಎಸ್ ಸಂಗೂರ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644236", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009377.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 32, \"b_dob\": \"2004-03-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹುಸೇನಮಿಯಾ\", \"b_father_kn\": \"ಜಾಫರ್ ಸಾಬ ಮುದೂರ\", \"mother_namekn\": \"ಅಕೈಲಾ\", \"info_name_kn\": \"ಗಾಯತ್ರಿ ಕೆ\", \"info_details\": \"Anganawadi Karyakarthe Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009377", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009371.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 26, \"b_dob\": \"2004-02-22\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಅಲ್ತಾಫ್ ಅಹಮದ್\", \"mother_namekn\": \"ಜಾಹಿರುನ್ನೀಸಾ\", \"info_name_kn\": \"ಎಂ ಓ ಹೆಚ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009371", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048548.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 13, \"b_dob\": \"1969-05-16\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸರೋಜಿ\", \"b_father_kn\": \"ಪಕ್ಕೀರಪ್ಪ ಹನುಮಂತಪ್ಪ ಲಕ್ಷ್ಮೀಶ್ವರ\", \"mother_namekn\": \"ಪಾರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048548", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009363.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 18, \"b_dob\": \"2004-02-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬೀರೇಶ್\", \"b_father_kn\": \"ಶಿವಪ್ಪ ಗುಡಗುಡಿ\", \"mother_namekn\": \"ಕಸ್ತೂರಿ\", \"info_name_kn\": \"ಸಿ ವಿ ಶರವಣ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009363", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047274.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 39, \"b_dob\": \"1958-12-02\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಿಂಗಪ್ಪ\", \"b_father_kn\": \"ಕರಿಯಲ್ಲಪ್ಪ ಪಕ್ಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 40, \"b_dob\": \"1958-12-04\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಹನಮಪ್ಪ ದುರಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 41, \"b_dob\": \"1958-12-18\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಬಸಪ್ಪ ಸಿದ್ದಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 42, \"b_dob\": \"1958-12-20\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನಮಪ್ಪ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 43, \"b_dob\": \"1958-12-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಸಿವಪ್ಪ ಬಸವಣ್ಣಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 44, \"b_dob\": \"1958-12-28\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹೊಳಿಬಸಪ್ಪ\", \"b_father_kn\": \"ಇರಪ್ಪ ಚನವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 6, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047274", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009358.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 13, \"b_dob\": \"2004-02-07\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಮಹ್ಮದಲಿ ಅಬ್ದುಲ್ ಖಾದರ್ ಬಾಳೂರು\", \"mother_namekn\": \"ರೇಷ್ಮಾ ಮಹ್ಮದಲಿ ಬಾಳೂರು\", \"info_name_kn\": \"ಪರಮೇಶ್ವರಿ ಎಸ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009358", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/BELAKERI/B_540_5493_046753.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 21, \"b_dob\": \"1958-06-14\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಫಕೀರಪ್ಪ\", \"b_father_kn\": \"ಕರಲಿಂಗಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1958-06-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗಂಗಾಮಲವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಚನ್ನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 23, \"b_dob\": \"1958-06-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಈಶ್ವರಪ್ಪ\", \"b_father_kn\": \"ಭದ್ರಪ್ಪ ತಿರಕಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 24, \"b_dob\": \"1958-06-21\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಮಾಳಪ್ಪ\", \"mother_namekn\": \"ಬೆಳಕೇರಿ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 28, \"b_dob\": \"1958-08-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕಾಳಪ್ಪ\", \"b_father_kn\": \"ಶಿದಪ್ಪ ಈರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 29, \"b_dob\": \"1958-08-21\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕಮಲವ್ವ\", \"b_father_kn\": \"ಈರನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 30, \"b_dob\": \"1958-06-11\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಫಕೀರವ್ವ\", \"b_father_kn\": \"ಮುದುಕಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_046753", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009346.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"2004-01-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಖಮರಾ ಅಲಿ\", \"b_father_kn\": \"ದುಡ್ಡು ಬಸೀರಾ ನಾಯಕ್\", \"mother_namekn\": \"ಮಹಬೂಬಾಬಿ\", \"info_name_kn\": \"ದುಡ್ಡುವಾ ಬಸೇರ ನಾಯಕ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009346", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053558.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 0, \"b_dob\": \"1958-02-28\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟವ್ವ\", \"b_father_kn\": \"ಭೀರಪ್ಪ ಚನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 27, \"b_dob\": \"1958-03-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಉಪಸಯ್ಯ\", \"b_father_kn\": \"ಲಕ್ಷ್ಮಪ್ಪ ದುರ್ಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 28, \"b_dob\": \"1958-03-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರಾಚಪ್ಪ\", \"b_father_kn\": \"ಕರಬಸಪ್ಪ ರಾಚಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 29, \"b_dob\": \"1958-03-06\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಪ್ಪ\", \"b_father_kn\": \"ಚಂಪ ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 30, \"b_dob\": \"1958-03-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಫಕೀರವ್ವ\", \"b_father_kn\": \"ನಾಗಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 31, \"b_dob\": \"1958-03-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರಾಮಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಮಾಳಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 32, \"b_dob\": \"1958-03-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಲಕ್ಷ್ಮವ್ವ\", \"b_father_kn\": \"ರಂಗಪ್ಪ ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 33, \"b_dob\": \"1958-03-15\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಾಂತವ್ವ\", \"b_father_kn\": \"ಮಹದೇವಪ್ಪ ಚನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 8, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053558", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009375.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 30, \"b_dob\": \"2004-02-26\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಹಜರತಾಳಿ ಹೊಸೂರ\", \"mother_namekn\": \"ಅಕಾಲಿಮಾ\", \"info_name_kn\": \"MOH PHC\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009375", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009368.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 23, \"b_dob\": \"2004-02-20\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಇಬ್ರಾಹಿಂ\", \"b_father_kn\": \"ಜೈರುದ್ದೀನ್ ಫಕ್ಕೀರಸಾಬಣ್ಣವರ\", \"mother_namekn\": \"ರೇಷ್ಮಾ\", \"info_name_kn\": \"ಗಾಯತ್ರಿ ಕೆ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009368", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/KALAGONDA/D_540_5493_644259.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 5, \"d_dod\": \"2012-12-08\", \"gender\": \"F\", \"age\": 70, \"d_deceased_nme_kn\": \"ಪದ್ಮವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಪದಮಪ್ಪ ಮುಳಿಕೇರಿ\", \"mother_namekn\": null, \"info_name_kn\": \"ಪದಮಪ್ಪ ಶಾಂತಪ್ಪ ಮಾಳಿಕೇರಿ\", \"info_details\": \"Malikeri\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644259", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048566.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 32, \"b_dob\": \"1970-12-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸುಭಾಸು\", \"b_father_kn\": \"ಬಸವಣ್ಣೆಪ್ಪ ಗುತ್ತೆಪ್ಪ ಹಂಚಿನಮನಿ\", \"mother_namekn\": \"ಶ್ಯಾಂತವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 33, \"b_dob\": \"1970-12-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮರಿಯವ್ವ\", \"b_father_kn\": \"ಮುದ್ದಿ ಮಲ್ಲಪ್ಪ ನೀಲಪ್ಪ\", \"mother_namekn\": \"ಭರಮವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 34, \"b_dob\": \"1970-12-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬೀರಪ್ಪ\", \"b_father_kn\": \"ತಿರಕಪ್ಪ ರಾಮನಹಳ್ಳಿ\", \"mother_namekn\": \"ಪುಟ್ಟವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 35, \"b_dob\": \"1970-12-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪ್ರೇಮಿಲಾ\", \"b_father_kn\": \"ಪರಸಪ್ಪ ಬಸವಣ್ಣೆಪ್ಪ ಮಡಿವಾಳರ\", \"mother_namekn\": \"ಸುಂದ್ರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 36, \"b_dob\": \"1970-12-11\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಜೀವಪ್ಪ ಮಾರ್ತಾಂಡಪ್ಪ ಕಾಟಿ\", \"mother_namekn\": \"ತಿಪ್ಪವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 37, \"b_dob\": \"1970-12-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬರಮಪ್ಪ\", \"b_father_kn\": \"ಬಸಟ್ಟೆಪ್ಪ ಬರಮಪ್ಪ\", \"mother_namekn\": \"ನೀಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 38, \"b_dob\": \"1971-01-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ತಿರಕಪ್ಪ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಭರಮಪ್ಪ ಗೌಡ್ರ\", \"mother_namekn\": \"ನೀಲಮ್ಮ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048566", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/KALAGONDA/D_540_5493_644257.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 3, \"d_dod\": \"2012-10-29\", \"gender\": \"M\", \"age\": 70, \"d_deceased_nme_kn\": \"ದುರಗಪ್ಪ ನಾಗಪ್ಪ ಹರಿಜನ\", \"d_father_kn\": \"ನಾಗಪ್ಪ ಹರಿಜನ ಊರ್ಫ್ ಮಾದರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಇಂದ್ರವ್ವ\", \"info_details\": \"Mallapura\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644257", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/BELAKERI/B_540_5493_046751.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 10, \"b_dob\": \"1958-01-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಲ್ಲವ್ವ\", \"b_father_kn\": \"ಶಿವಯ್ಯ ವೀರಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 11, \"b_dob\": \"1958-03-28\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಾಗವ್ವ\", \"b_father_kn\": \"ತಿರಕನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 12, \"b_dob\": \"1958-04-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಬಸವ್ವ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ ಹೊನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 13, \"b_dob\": \"1958-04-11\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಬಸಪ್ಪ ನಾಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 14, \"b_dob\": \"1958-04-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗುಡಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ನೀಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 15, \"b_dob\": \"1958-04-16\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಿವನಗವ್ವ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ಈಶ್ವರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 16, \"b_dob\": \"1958-04-23\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಿವಕ್ಕ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಆದಿಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 17, \"b_dob\": \"1958-05-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚಾನವ್ವ\", \"b_father_kn\": \"ಉಜಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 8, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_046751", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/DUMMIHALA/D_540_5493_644173.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 4, \"d_dod\": \"2012-03-12\", \"gender\": \"M\", \"age\": 48, \"d_deceased_nme_kn\": \"ಬಸುರಾಜ ಶೇಡಿಯಪ್ಪ ರಾಮಕ್ಕನವರ\", \"d_father_kn\": null, \"d_h_w_name_kn\": null, \"mother_namekn\": \"ಹನುಮವ್ವ\", \"info_name_kn\": \"ರಮೇಶ ಎಸ್ ರಾಮಕ್ಕನವರ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644173", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/BELAKERI/B_540_5493_046754.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 25, \"b_dob\": \"1958-10-02\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗೌರವ್ವ\", \"b_father_kn\": \"ಭರಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 35, \"b_dob\": \"1958-10-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪರವತಪ್ಪ\", \"b_father_kn\": \"ಚನಬಸಪ್ಪ ಕರೇಗೌಡ ಡಾ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 36, \"b_dob\": \"1958-10-12\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪರವತವ್ವ\", \"b_father_kn\": \"ಹೊಳಿಬಸಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 37, \"b_dob\": \"1958-10-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರೇವಣಪ್ಪ\", \"b_father_kn\": \"ಚನ್ನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 38, \"b_dob\": \"1958-10-30\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮದರಸಾಬ\", \"b_father_kn\": \"ಜಮಲಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 39, \"b_dob\": \"1958-11-02\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತಪ್ಪ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ಹೋಬನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 40, \"b_dob\": \"1958-11-14\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಈರಪ್ಪ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 41, \"b_dob\": \"1958-11-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರೇಣುಕಾ\", \"b_father_kn\": \"ಸಣ್ಣಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 42, \"b_dob\": \"1958-11-22\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಬಸಪ್ಪ ಹನುಮಂತಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 9, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_046754", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053559.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 34, \"b_dob\": \"1958-03-15\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ತಿರಕಪ್ಪ\", \"b_father_kn\": \"ಪುಟ್ಟಪ್ಪ ಚನ್ನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 35, \"b_dob\": \"1958-03-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟವ್ವ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ನಾಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 36, \"b_dob\": \"1958-03-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗಂಗಪ್ಪ\", \"b_father_kn\": \"ದೇವನಗೌಡ ಗುರುಪಾದಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 37, \"b_dob\": \"1958-03-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟವ್ವ\", \"b_father_kn\": \"ರಾಮನಗೌಡ ಭೀಮನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 38, \"b_dob\": \"1958-03-23\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪರವತವ್ವ\", \"b_father_kn\": \"ರಾಮಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 39, \"b_dob\": \"1958-03-24\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವರಾಜು\", \"b_father_kn\": \"ರಾಮಪ್ಪ ಗುಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 40, \"b_dob\": \"1958-03-28\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವರಾಜು\", \"b_father_kn\": \"ಬಸವಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 41, \"b_dob\": \"1958-04-04\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಬಸವರಾಜು\", \"b_father_kn\": \"ಚನ್ನಬಸಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 8, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053559", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053564.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-06-16\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಮಾಳಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-06-16\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರತ್ನವ್ವ\", \"b_father_kn\": \"ಮಾನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 3, \"b_dob\": \"1958-06-18\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಾವಂತರವ್ವ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-06-25\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಾವಂತರವ್ವ\", \"b_father_kn\": \"ಕಾಳಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1958-06-25\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದುರ್ಗವ್ವ\", \"b_father_kn\": \"ಮಡಿವಾಳಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1958-06-25\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ವಿನೋದಾಬಾಯಿ\", \"b_father_kn\": \"ಮಲಾರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 87, \"b_dob\": \"1958-06-26\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗಿಡಬಸಪ್ಪ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 88, \"b_dob\": \"1958-07-03\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕಸ್ತೂರಿ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 89, \"b_dob\": \"1958-07-09\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಫಕೀರಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 9, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053564", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/BELAKERI/B_540_5493_0046750.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-11-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಿವಪ್ಪ\", \"b_father_kn\": \"ಹೊನಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-10-20\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸುಶೀಲವ್ವ\", \"b_father_kn\": \"ಅಲ್ಲಾಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 3, \"b_dob\": \"1958-10-15\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸ್ವಾಮಿಲಿಂಗಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಕರಿಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-02-13\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಿವಕ್ಕ\", \"b_father_kn\": \"ಭರಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1958-03-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸೀತವ್ವ\", \"b_father_kn\": \"ಶಿವಪ್ಪ ನೆಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1958-03-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರತ್ನಾಕರ\", \"b_father_kn\": \"ರವಿಚಂದ್ರಪ್ಪ ಗೋವಿಂದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1958-03-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರತ್ನಾಕರ\", \"b_father_kn\": \"ರಾಮಚಂದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 8, \"b_dob\": \"1958-03-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪಿರಕಪ್ಪ\", \"b_father_kn\": \"ರುದ್ರಯ್ಯ ವೀರಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 9, \"b_dob\": \"1958-03-17\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಬಸಪ್ಪ ಭೀಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 9, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0046750", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MALLAURU/B_540_5493_0060777.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-01-04\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಂಕರಪ್ಪ\", \"b_father_kn\": \"ಮಾಳಪ್ಪ ಪರಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-01-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಂಕ್ರವ್ವ\", \"b_father_kn\": \"ಕೊತ್ರಯ್ಯ ವೀರನಿಂಗಯ್ಯ\", \"mother_namekn\": \"ಮಲ್ಲೂರು\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 3, \"b_dob\": \"1958-01-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಜಗಜನನಿ\", \"b_father_kn\": \"ಶ್ರೀಭೂಷಣಸ್ವಾಮಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-11-24\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಂಜುಳಾ ಬಾಯಿ\", \"b_father_kn\": \"ತಿಪಣ್ಣ ಮಾನಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1958-02-09\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ತಿರಕಪ್ಪ\", \"b_father_kn\": \"ಶಿವಪ್ಪ ತಿರಕಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1958-02-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಿವಗಂಗವ್ವ\", \"b_father_kn\": \"ನಾಗಪ್ಪ ವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1958-02-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹಾಭೂಬಲಿ\", \"b_father_kn\": \"ಮಾಮನಸಾಬ ಉಸೇನಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 8, \"b_dob\": \"1958-02-20\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಜೋರಾಬಿ\", \"b_father_kn\": \"ಇಮಾಮಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 9, \"b_dob\": \"1958-02-24\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಲೈಕರಾಜುನಪ್ಪ\", \"b_father_kn\": \"ಮಲಾರಪ್ಪ ನಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 10, \"b_dob\": \"1958-01-13\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟನಗೌಡ\", \"b_father_kn\": \"ತಿರಕನಗೌಡ ವೀರಭದ್ರಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 11, \"b_dob\": \"1958-03-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ವೀರಪ್ಪ\", \"b_father_kn\": \"ವೀರಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 12, \"b_dob\": \"1958-03-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಹದೇವಕ್ಕ\", \"b_father_kn\": \"ವೀರಭದ್ರಪ್ಪ ಸೋಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 13, \"b_dob\": \"1958-03-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ವೀರವ್ವ\", \"b_father_kn\": \"ಚಂದ್ರಶೇಖರಪ್ಪ ನಾಗಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 14, \"b_dob\": \"1958-03-25\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರೇಣುಕಾಬಾಯಿ\", \"b_father_kn\": \"ತಿಮಣ್ಣ ನರಸಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 14, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0060777", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048552.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 21, \"b_dob\": \"1969-08-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಜೀರಸಾಬ\", \"b_father_kn\": \"ಪೀರಸಾಬ ರಾಜೇಸಾಬ ಹಾವೇರಿ\", \"mother_namekn\": \"ಬಡಿಬು\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1969-09-02\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚನ್ನಪ್ಪ\", \"b_father_kn\": \"ಸಂಕಪ್ಪ ಹನುಮಣ್ಣ ಪೂಜಾರ\", \"mother_namekn\": \"ನೀಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048552", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009364.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 19, \"b_dob\": \"2004-02-13\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಜಬೀವುಲ್ಲಾ\", \"b_father_kn\": \"ನಜೀರ ಅಹಮದ್ ಸೋಮಸಾಗರ\", \"mother_namekn\": \"ಹಸೀನಾ\", \"info_name_kn\": \"ಸಿ ವಿ ಶರವಣ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009364", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644243.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 12, \"d_dod\": \"2012-06-10\", \"gender\": \"M\", \"age\": 30, \"d_deceased_nme_kn\": \"ಶಿವಪ್ಪ ಬಸಪ್ಪ\", \"d_father_kn\": \"ಬಸಪ್ಪ ಸುಬ್ರಗಟ್ಟಿ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ತಿರಕಪ್ಪ ನಾಗಪ್ಪ ಮುಳಗುಂದ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644243", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048551.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 17, \"b_dob\": \"1969-08-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತಪ್ಪ\", \"b_father_kn\": \"ಮುದ್ದೆಪ್ಪ ಹನುಮಂತಪ್ಪ ಹುಲಿಹಳ್ಳಿ\", \"mother_namekn\": \"ಪಾರ್ವತೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 18, \"b_dob\": \"1969-08-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತಪ್ಪ\", \"b_father_kn\": \"ಪಕ್ಕೀರಪ್ಪ ಮಲ್ಲಪಕನವಳ್ಳಿ\", \"mother_namekn\": \"ಹನುಮವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 19, \"b_dob\": \"1969-08-02\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಹನುಮಂತಗೌಡ ಖಾನಗೌಡ ಪಾಟೀಲ\", \"mother_namekn\": \"ಬಸವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 20, \"b_dob\": \"1969-08-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಲಕ್ಷಮವ್ವ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಗುಡ್ಡಪ್ಪ ಬುಳ್ಳನವರ\", \"mother_namekn\": \"ಲಕ್ಷ್ಮವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048551", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048562.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 20, \"b_dob\": \"1970-08-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಿರ್ಮಲಾ\", \"b_father_kn\": \"ಗುಡ್ಡಪ್ಪ ಚಿಕ್ಕಪ್ಪ ಬಿಲ್ಲಳ್ಳಿ\", \"mother_namekn\": \"ಯಲ್ಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 21, \"b_dob\": \"1970-08-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಅಬ್ದುಲ್ ರಹಿಮಾನ್\", \"b_father_kn\": \"ಗೌಸ ಮೋದಿನ ಸಾಹೇಬ ಮೋದಿನಾಸಾಹೇಬ ವಾಲಿಕಾರ\", \"mother_namekn\": \"ವಜೀರಿಬಿ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1970-08-27\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ನೀಲಪ್ಪ ನಾಗಪ್ಪ ಕಳ್ಳಿಮನಿ\", \"mother_namekn\": \"ಶಾರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 23, \"b_dob\": \"1970-08-27\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಾದೇವಕ್ಕ\", \"b_father_kn\": \"ಹನುಮಂತಪ್ಪ ರಿತ್ತಿ\", \"mother_namekn\": \"ದುರುಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048562", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/KENGONDA/B_540_5493_061668.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 6, \"b_dob\": \"1958-02-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಿಂಗವ್ವ\", \"b_father_kn\": \"ವೀರುಪಾಕ್ಷಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1958-03-15\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 8, \"b_dob\": \"1958-03-25\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಭೀಮವ್ವ\", \"b_father_kn\": \"ಭೀಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 9, \"b_dob\": \"1958-04-13\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಅಬ್ದುಲಸಾಬ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 10, \"b_dob\": \"1958-05-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಲ್ಲವ್ವ\", \"b_father_kn\": \"ವಿನಂದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 11, \"b_dob\": \"1958-05-21\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟವ್ವ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 6, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_061668", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MALLAURU/B_540_5493_060778.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 15, \"b_dob\": \"1958-01-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರೇಣುಕಾ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ ರಾಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 16, \"b_dob\": \"1958-01-13\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗೌರವ್ವ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ಹುಚ್ಚಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 17, \"b_dob\": \"1958-01-25\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಅಕ್ಕಮಹಾದೇವಿ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಕರಿಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 18, \"b_dob\": \"1958-05-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಅಮಿನ್ಬಿ\", \"b_father_kn\": \"ರಾಜೇಸಾಬ ಜಂಗಲಿಸಾಬ್\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 19, \"b_dob\": \"1958-05-12\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರತ್ನವ್ವ\", \"b_father_kn\": \"ಜೋತಪ್ಪ ರಾಜಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 20, \"b_dob\": \"1958-05-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚಮ್ನಸಾಬ\", \"b_father_kn\": \"ಸಣ್ಣರಾಜೇಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 21, \"b_dob\": \"1958-05-28\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ರಾಜಾಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1958-01-15\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಅನ್ವರಳ್ಳಿ\", \"b_father_kn\": \"ಚಮನಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 23, \"b_dob\": \"1958-06-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಾನಬಸಪ್ಪ\", \"b_father_kn\": \"ಆದಿನಪ್ಪ ಭೀಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 24, \"b_dob\": \"1958-06-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕುಮಾರಸ್ವಾಮಿ\", \"b_father_kn\": \"ಶಿವಯ್ಯ ಮುನಿಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 25, \"b_dob\": \"1958-06-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಲ್ಲಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಭರಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 26, \"b_dob\": \"1958-06-20\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಈರವ್ವ\", \"b_father_kn\": \"ವೀರಯ್ಯ ಚಂದ್ರಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 27, \"b_dob\": \"1958-06-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಾಂತವ್ವ\", \"b_father_kn\": \"ಕರಿಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 28, \"b_dob\": \"1958-06-24\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಅನ್ನಪೂರ್ಣ\", \"b_father_kn\": \"ಕಲವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 29, \"b_dob\": \"1958-06-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹುಚ್ಚಪ್ಪ\", \"b_father_kn\": \"ದುರ್ಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 30, \"b_dob\": \"1958-07-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಗುಡಪ್ಪ ನೀಲಪ್ಪ ಭಾಂಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 31, \"b_dob\": \"1958-07-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪಾರವ್ತವ್ವ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ಮರಿಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 32, \"b_dob\": \"1958-07-07\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗುಲಾರಸಾಬ\", \"b_father_kn\": \"ಪಿರಾಮಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 33, \"b_dob\": \"1958-07-02\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಲ್ಲವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಬಸಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 34, \"b_dob\": \"1958-07-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ತಿಮಣ್ಣ\", \"b_father_kn\": \"ಸುಭಣ್ಣ ಪಿರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 35, \"b_dob\": \"1958-07-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಫಕೀರಪ್ಪ\", \"b_father_kn\": \"ಈರಪ್ಪ ಗುರುಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 21, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_060778", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048561.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 16, \"b_dob\": \"1970-07-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ತಿರಕವ್ವ\", \"b_father_kn\": \"ಗುಡ್ಡಪ್ಪ ನಿಂಗಪ್ಪ ಚೋಳಕೇರಿ\", \"mother_namekn\": \"ಚಿನ್ನವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 17, \"b_dob\": \"1970-07-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ನಿಂಗನಗೌಡ ನೀಲನಗೌಡ\", \"mother_namekn\": \"ಮಾಳವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 18, \"b_dob\": \"1970-07-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶೇಕಪ್ಪ\", \"b_father_kn\": \"ತಿರಕಪ್ಪ ನಿಂಗಪ್ಪ ಗಾಜಿ\", \"mother_namekn\": \"ಪಾರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 19, \"b_dob\": \"1970-07-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಾಹೇಬಿ\", \"b_father_kn\": \"ಪೀರಮಮ್ಮದ ಸಾಬ ಮದರಸಾಬ ಹಾವೇರಿ\", \"mother_namekn\": \"ಜೈತುನಬಿ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048561", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/GUDDADAMALLAPURA/D_540_5493_644171.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 2, \"d_dod\": \"1996-03-06\", \"gender\": \"F\", \"age\": 65, \"d_deceased_nme_kn\": \"ಬಸವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಯಲ್ಲಪ್ಪ ಸೋದ್ರಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": \"Guddada Mallapur\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644171", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009379.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 34, \"b_dob\": \"2004-03-09\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಬಾಫ್ರೀನಾ\", \"b_father_kn\": \"ಅಹ್ಮದ್ ಖಾದರ್ ಶಿಡಗೇದ\", \"mother_namekn\": \"ಕೌಸರ ಜಮ್ಹಾ\", \"info_name_kn\": \"ಶಹತಾಜ್ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009379", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009370.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 25, \"b_dob\": \"2004-02-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಹಿದಾ ಉರುಫ್ ನೂರ್ ಅಪ್ಸಯಾ\", \"b_father_kn\": \"ಸೈಫೀರ್ ಶೇತ್ಸನದಿ\", \"mother_namekn\": \"ಶೇಕಿಲಾ ಸೈಫೀರ್ ಶೇಕಸಂಧಿ\", \"info_name_kn\": \"ಪಿ ಎಸ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009370", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644238.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 17, \"d_dod\": \"2012-09-13\", \"gender\": \"M\", \"age\": 68, \"d_deceased_nme_kn\": \"ಗುಡ್ಡಪ್ಪ ಮಾರ್ತಾಂಡಪ್ಪ ಬಡಿಗೇರ\", \"d_father_kn\": \"ಮಾರ್ತಾಂಡಪ್ಪ ಬಡಿಗೇರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಶ್ರೇಣಿಕರಾಜ ಎಸ್ ಯಲವತ್ತಿ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644238", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644240.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 15, \"d_dod\": \"1982-08-12\", \"gender\": \"M\", \"age\": 30, \"d_deceased_nme_kn\": \"ಸೋಮಪ್ಪ ಚವಟಿ\", \"d_father_kn\": \"ಹನುಮಪ್ಪ ಚವಟಿ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಸೋಮವ್ವ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644240", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_048018.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 39, \"b_dob\": \"1958-11-27\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 40, \"b_dob\": \"1958-12-04\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಯಲ್ಲಪ್ಪ\", \"b_father_kn\": \"ಹನುಮಂತಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 41, \"b_dob\": \"1958-12-20\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚವಡಪ್ಪ\", \"b_father_kn\": \"ಮೂಕಪ್ಪ ಮರಿಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 42, \"b_dob\": \"1958-12-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮೂಕಬಸಪ್ಪ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ಸಿದ್ದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 43, \"b_dob\": \"1958-12-24\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಂಕರಗೌಡ\", \"b_father_kn\": \"ಭರಮಪ್ಪ ನಾಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 44, \"b_dob\": \"1958-12-26\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಹಾಲಪ್ಪ ಬಾಳಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 6, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048018", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009373.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 28, \"b_dob\": \"2004-02-23\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಇಮಾಮಾ ಹುಸೇನ್\", \"b_father_kn\": \"ಅನ್ವರ್ ಬಾಶ್ಯ ಅತ್ತರ್\", \"mother_namekn\": \"ಫತೀಮಾಬಿ\", \"info_name_kn\": \"ಅನ್ವರ್ ಬಾಶ್ಯ ಅತ್ತರ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009373", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048547.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 7, \"b_dob\": \"1969-04-04\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸುಶೀಲವ್ವ\", \"b_father_kn\": \"ತಿಪ್ಪಣ್ಣ ಕಲ್ಲೇದೇವರು\", \"mother_namekn\": \"ಯಲ್ಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 8, \"b_dob\": \"1969-04-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಪರಸಪ್ಪ ಬಸಪ್ಪ ಮಡಿವಾಳರ\", \"mother_namekn\": \"ಸುಂದ್ರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 9, \"b_dob\": \"1969-04-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಚಿಕ್ಕಪ್ಪ ನಾಗಪ್ಪ ಮಾತವರ\", \"mother_namekn\": \"ನೀಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 10, \"b_dob\": \"1969-04-11\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಬಸುಬಿ\", \"b_father_kn\": \"ನನ್ನೇಸಾಬ ಜಮಾಲ ಸಾಬ ವಾಲೀಕಾರ\", \"mother_namekn\": \"ಹುಶೇನ್ಬಿ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 11, \"b_dob\": \"1969-04-14\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವಣ್ಣೆಪ್ಪ\", \"b_father_kn\": \"ಎಲ್ಲಪ್ಪ ಗುಡ್ಡಪ್ಪ ಗಾಜಿ\", \"mother_namekn\": \"ಗಂಗಿಮಾಳವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 12, \"b_dob\": \"1969-04-30\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸುರೇಶ\", \"b_father_kn\": \"ಗುಡ್ಡಪ್ಪ ಬೀರಪ್ಪ ಬಣಕಾರ\", \"mother_namekn\": \"ಗಂಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 6, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048547", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/BELAKERI/B_540_5493_046755.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 43, \"b_dob\": \"1958-12-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗುಡಪ್ಪ\", \"b_father_kn\": \"ಚಿಕ್ಕಪ್ಪ ಗುಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 44, \"b_dob\": \"1958-03-17\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗುಡಪ್ಪ\", \"b_father_kn\": \"ಚಿಕ್ಕಪ್ಪ ಗುಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_046755", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009352.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 7, \"b_dob\": \"2004-02-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹ್ಮದ್ ಅಫ್ತಾಬಾ\", \"b_father_kn\": \"ಇಮ್ತಿಯಾ ಅಹ್ಮದ್ ಮುಕಾಶಿ\", \"mother_namekn\": \"ಗುಲಾಬಾ ಮುಕಾಶಿ\", \"info_name_kn\": \"ಪರಮೇಶ್ವರಿ ಎಸ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009352", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053557.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-02-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತ\", \"b_father_kn\": \"ಭೀಮಪ್ಪ ರಾಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-02-02\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಲ್ಲಯ್ಯ\", \"b_father_kn\": \"ವೀರಪಕ್ಷಯ್ಯ ಗುರುಬಸಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 0, \"b_dob\": \"1958-02-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ತಿರಕಪ್ಪ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-02-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಅಲ್ಲಾಮೋದಿನಾ\", \"b_father_kn\": \"ಹುಸೇನಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1958-02-07\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಿಂಗವ್ವ\", \"b_father_kn\": \"ನೀಲಪ್ಪ ದುರ್ಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1958-02-08\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ನೀಲಪ್ಪ ಚನ್ನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1958-02-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ದೇವಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ದೇವಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053557", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048550.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 15, \"b_dob\": \"1969-07-04\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗಂಗಿಮಾಳವ್ವ\", \"b_father_kn\": \"ಗುಡ್ಡಪ್ಪ ಪುಟ್ಟಪ್ಪ ಬಣಕಾರ\", \"mother_namekn\": \"ಮಲ್ಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 16, \"b_dob\": \"1969-07-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ನೀಲಪ್ಪ ಕೋಡಿಹಳ್ಳಿ\", \"mother_namekn\": \"ಗಂಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048550", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048565.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 29, \"b_dob\": \"1970-11-12\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹೊನ್ನವ್ವ\", \"b_father_kn\": \"ಶಿವಪ್ಪ ಭೀಮಪ್ಪ ಹಾವೇರಿ\", \"mother_namekn\": \"ಪಾರ್ವತೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 30, \"b_dob\": \"1970-11-19\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರೇಣುಕಾ\", \"b_father_kn\": \"ಸಣ್ಣ ಗುಡ್ಡಪ್ಪ ಹೊನ್ನಪ್ಪ ಗಾಜಿ\", \"mother_namekn\": \"ಅಡಿವಕ್ಕ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 31, \"b_dob\": \"1970-11-19\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸರಸ್ವತೆವ್ವ\", \"b_father_kn\": \"ಗುಡ್ಡನ ಗೌಡ ನೀಲನಗೌಡ ಪಾಟೀಲ\", \"mother_namekn\": \"ಗೌರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 3, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048565", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009374.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 29, \"b_dob\": \"2004-02-24\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಸುಲೇಮಾನ ಹಾವೇರಿ\", \"mother_namekn\": \"ಶಮಶಾದ್\", \"info_name_kn\": \"MOH PHC\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009374", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644244.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 11, \"d_dod\": \"1973-06-05\", \"gender\": \"F\", \"age\": 60, \"d_deceased_nme_kn\": \"ಭರಮವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಕರಿಯಪ್ಪ ಕಾಕೋಳ\", \"mother_namekn\": null, \"info_name_kn\": \"ದಿಲ್ಲೆಪ್ಪ ಹುಚ್ಚಪ್ಪ ಕಾಕೋಳ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644244", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047272.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 24, \"b_dob\": \"1958-08-02\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಸಿವರಾಮಪ್ಪ ಇರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 25, \"b_dob\": \"1958-09-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತಪ್ಪ\", \"b_father_kn\": \"ಇರಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 26, \"b_dob\": \"1958-01-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚೌಡವ್ವ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 27, \"b_dob\": \"1958-10-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಭೋಜಪ್ಪ\", \"b_father_kn\": \"ಭೀಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 28, \"b_dob\": \"1958-10-15\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಸಿವಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 29, \"b_dob\": \"1958-10-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗಂಗವ್ವ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 6, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047272", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009351.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 6, \"b_dob\": \"2004-02-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕಾಂತೇಶ\", \"b_father_kn\": \"ಎಲ್ಲಪ್ಪ ಈಳಿಗೇರ\", \"mother_namekn\": \"ನಾಗಮ್ಮ\", \"info_name_kn\": \"ಈರಪ್ಪ ತಳವಾರ\", \"info_details\": null}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009351", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009359.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 14, \"b_dob\": \"2004-02-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕಿರಣ\", \"b_father_kn\": \"ಆನಂದ ಕುರಗೋಡಿ\", \"mother_namekn\": \"ಗೀತಾ\", \"info_name_kn\": \"ಸಿ ವಿ ಶರವಣ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009359", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/KALAGONDA/D_540_5493_644255.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"d_dod\": \"2012-03-20\", \"gender\": \"M\", \"age\": 72, \"d_deceased_nme_kn\": \"ಮಲ್ಲಪ್ಪ\", \"d_father_kn\": \"ಹನುಮಪ್ಪ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಕಲಗೊಂಡ\", \"info_details\": \"Suresh Mallappa Talavar\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644255", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048553.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 23, \"b_dob\": \"1969-11-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಬಸವ್ವ\", \"b_father_kn\": \"ಮಡಿಗೌಡ ಗುಡ್ಡನಗೌಡ\", \"mother_namekn\": \"ರಿಂದವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048553", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048554.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 24, \"b_dob\": \"1969-12-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹಾದೇವಪ್ಪ\", \"b_father_kn\": \"ಶಿದ್ದಪ್ಪ ಎಲ್ಲಪ್ಪ\", \"mother_namekn\": \"ಪಾರ್ವತೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 25, \"b_dob\": \"1969-12-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರಾಮಣ್ಣ\", \"b_father_kn\": \"ತಿಪ್ಪಣ್ಣ ಬಸಪ್ಪ ಗುರಣ್ಣನವರ\", \"mother_namekn\": \"ಗುತ್ತೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 26, \"b_dob\": \"1969-12-26\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಜುಂಜನಗೌಡ\", \"b_father_kn\": \"ಬಸನಗೌಡ ಭರಮ ಗೌಡ ಗೌಡ್ರ\", \"mother_namekn\": \"ನಾಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 3, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048554", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009349.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 4, \"b_dob\": \"2004-01-17\", \"gender\": \"F\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಅರ್ವಮಾನ ಸಾಬ\", \"mother_namekn\": \"ರಜೀಯಾ ಬಾನು\", \"info_name_kn\": \"ಶಹತಾಜ್ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009349", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/GUDDADAMALLAPURA/D_540_5493_644158.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 3, \"d_dod\": \"2012-08-12\", \"gender\": \"F\", \"age\": 118, \"d_deceased_nme_kn\": \"ಫೀರಂಬಿ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಫೀರು ಸಾಬ್ ಉಪ್ಪುಣಸಿ\", \"mother_namekn\": null, \"info_name_kn\": \"ಜಹೇದಾ\", \"info_details\": \"Guddada Mallapur\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644158", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/TIPALAPURA/D_540_5493_644130.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"d_dod\": \"2012-01-18\", \"gender\": \"M\", \"age\": 42, \"d_deceased_nme_kn\": \"ದಾನಗೌಡ ಯಲ್ಲನಗೌಡ ಪಾಟೀಲ\", \"d_father_kn\": \"ಯಲ್ಲನಗೌಡ ಪಾಟೀಲ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಕರಬಸನಗೌಡ ದಾನಗೌಡ ಪಾಟೀಲ\", \"info_details\": \"Thippalapur\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644130", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048549.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 14, \"b_dob\": \"1969-06-19\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಹನುಮಂತಪ್ಪ ಬಸಪ್ಪ ಪೂಜಾರ\", \"mother_namekn\": \"ಚಿನ್ನವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048549", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644248.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 7, \"d_dod\": \"2012-04-20\", \"gender\": \"M\", \"age\": 65, \"d_deceased_nme_kn\": \"ಕಲವೀರಪ್ಪ ವೀರಪ್ಪ ಬಡಿಗೇರ\", \"d_father_kn\": \"ವೀರಪ್ಪ ಬಡಿಗೇರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಅನಿಲ ಕುಮಾರ ಕಲವೇರಪ್ಪ ಬಡಿಗೇರ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644248", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644249.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 6, \"d_dod\": \"2012-04-01\", \"gender\": \"F\", \"age\": 75, \"d_deceased_nme_kn\": \"ಸುಂದ್ರವ್ವ\", \"d_father_kn\": \"ರಾಮಪ್ಪ ಬಡಿಗೇರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಗಣೇಶಪ್ಪ ರಾಮಪ್ಪ ಬಡಿಗೇರ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644249", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644253.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 2, \"d_dod\": \"1977-12-05\", \"gender\": \"M\", \"age\": 70, \"d_deceased_nme_kn\": \"ಪುಟ್ಟಪ್ಪ ಸಣ್ಣ ಸೋಮಪ್ಪ ಹಿರೇಗೌಡ್ರ\", \"d_father_kn\": \"ಸಣ್ಣ ಸೋಮಪ್ಪ ಹಿರೇಗೌಡ್ರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಪುಟ್ಟಪ್ಪ ದಿಲ್ಲೆಪ್ಪ ಹಿರೇಗೌಡ್ರ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644253", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_0053556.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-01-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸರೋಜವ್ವ\", \"b_father_kn\": \"ಸಣ್ಣಹನುಮಂತಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-01-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗಂಗವ್ವ\", \"b_father_kn\": \"ಸಂಕಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 3, \"b_dob\": \"1958-01-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಾಗಪ್ಪ\", \"b_father_kn\": \"ಚನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-01-20\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹನುಮವ್ವ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ ಯಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1958-01-23\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಈರವ್ವ\", \"b_father_kn\": \"ಶಿವಪ್ಪ ಮಲಕಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1958-01-23\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಈರವ್ವ\", \"b_father_kn\": \"ಶಿಡಪ್ಪ ಮೂಕಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 7, \"b_dob\": \"1958-01-24\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸಪ್ಪ\", \"b_father_kn\": \"ಭರಮಪ್ಪ ಬಸಪ್ಪ ಮೈಲಾರ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 8, \"b_dob\": \"1958-01-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರತ್ನವ್ವ\", \"b_father_kn\": \"ಪೇತ್ರಪ್ಪ ಬಸವಂತಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 8, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0053556", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048564.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 28, \"b_dob\": \"1970-10-21\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತ\", \"b_father_kn\": \"ಅಂಬಾಡೆಪ್ಪ ಪಕ್ಕೀರಪ್ಪ ಕರಡೇರ\", \"mother_namekn\": \"ನಿಂಬವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 28, \"b_dob\": \"1970-10-28\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಾದೇವಪ್ಪ\", \"b_father_kn\": \"ಮಲ್ಲನಗೌಡ ಪಕ್ಕೀರ ಗೌಡ ಪಾಟೀಲ\", \"mother_namekn\": \"ಗೌರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048564", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048546.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 3, \"b_dob\": \"1969-02-09\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹನುಮಂತ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ನಿಂಗಪ್ಪ ಹಾವೇರಿ\", \"mother_namekn\": \"ಬಸವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1969-02-14\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಭರಮಪ್ಪ\", \"b_father_kn\": \"ಹನುಮಂತಪ್ಪ ಹೊನ್ನಪ್ಪ ವಾಲಿಕಾರ\", \"mother_namekn\": \"ಶ್ಯಾಂತವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1969-02-26\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಪಕ್ಕೀರಪ್ಪ ನಿಂಗಪ್ಪ ಹಾವೇರಿ\", \"mother_namekn\": \"ಪಾರ್ವತೆವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 6, \"b_dob\": \"1969-02-27\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನೀಲವ್ವ\", \"b_father_kn\": \"ಗಂಗಪ್ಪ ಚನಬಸಪ್ಪ ಬೂದನೂರ\", \"mother_namekn\": \"ಶೇಕ್ರವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1969\", \"total_records\": 4, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048546", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009362.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 17, \"b_dob\": \"2004-02-09\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮೈನುದ್ದೀನ್\", \"b_father_kn\": \"ಮಹ್ಮದ್ ರಫೀಕ್ ದಾರುಣವರ\", \"mother_namekn\": \"ಅಕೈಲಾ ಬಾನು\", \"info_name_kn\": \"ಪಿ ಎಸ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009362", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/KENGONDA/B_540_5493_061670.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 19, \"b_dob\": \"1958-10-20\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಹುಸೇನಗೌಡ\", \"b_father_kn\": \"ತಿರಕಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 20, \"b_dob\": \"1958-10-28\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ತಿರಕವ್ವ\", \"b_father_kn\": \"ದ್ಯಾಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 21, \"b_dob\": \"1958-11-28\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಾರುತಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಹಳಿಮುನಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1958-11-30\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಕೊತ್ತಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 23, \"b_dob\": \"1958-12-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ವೀರೇಂದ್ರಪ್ಪ\", \"b_father_kn\": \"ಕರಿಯಲ್ಲಾ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 24, \"b_dob\": \"1958-12-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದೇವಿಂದ್ರವ್ವ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 25, \"b_dob\": \"1958-12-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಲ್ಲನಗೌಡ\", \"b_father_kn\": \"ನಿಂಕನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_061670", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_0048017.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 32, \"b_dob\": \"1958-09-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹನುಮಂತವ್ವ\", \"b_father_kn\": \"ಸಾಕಪ್ಪ ಕರಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 33, \"b_dob\": \"1958-06-12\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹನುಮಂತವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 34, \"b_dob\": \"1958-09-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚೌಡಪ್ಪ\", \"b_father_kn\": \"ಕೆಂಚಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 35, \"b_dob\": \"1958-09-20\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹದೇವಪ್ಪ\", \"b_father_kn\": \"ಸಂಕಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 36, \"b_dob\": \"1958-09-23\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸಿದ್ದನಗೌಡ\", \"b_father_kn\": \"ಬಸನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 37, \"b_dob\": \"1958-10-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಲಕ್ಷ್ಮಪ್ಪ\", \"b_father_kn\": \"ಭರಮಪ್ಪ ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 38, \"b_dob\": \"1958-10-25\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕರಬಸಪ್ಪ\", \"b_father_kn\": \"ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0048017", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644246.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 9, \"d_dod\": \"2012-05-29\", \"gender\": \"F\", \"age\": 45, \"d_deceased_nme_kn\": \"ರೇಣಕವ್ವ\", \"d_father_kn\": \"ಬಸಪ್ಪ ಹಳೇಮನಿ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ನಾಗಪ್ಪ ಶಿವಪ್ಪ ಹಳೇಮನಿ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644246", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644239.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 16, \"d_dod\": \"1982-05-04\", \"gender\": \"F\", \"age\": 60, \"d_deceased_nme_kn\": \"ಸೋಮವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಸೋಮರಾಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": \"ಮಹಾಬಲೇಶ್ವರಪ್ಪ ಶೇಕಪ್ಪ ಪೂಜಾರ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644239", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048556.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 3, \"b_dob\": \"1970-02-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶ್ಯಾಂತವ್ವ\", \"b_father_kn\": \"ರಾಮನಗೌಡ ಹನುಮಂತಗೌಡ\", \"mother_namekn\": \"ಕರಿಲಿಂಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048556", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_048019.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 18, \"b_dob\": \"1900-01-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚನ್ನಬಸಪ್ಪ\", \"b_father_kn\": \"ಸಂಕಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 19, \"b_dob\": \"1958-06-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವಂತಪ್ಪ\", \"b_father_kn\": \"ಈರಪ್ಪ ಚನವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 20, \"b_dob\": \"1958-06-04\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚನಬಸಪ್ಪ\", \"b_father_kn\": \"ಈರಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 21, \"b_dob\": \"1958-06-06\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಜಂಬಪ್ಪ\", \"b_father_kn\": \"ಸಿದ್ದಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1958-06-07\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ ದುರುಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 23, \"b_dob\": \"1958-06-26\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರುಕ್ಮಣೆವ್ವ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 24, \"b_dob\": \"1958-07-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹನುಮವ್ವ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048019", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/DUMMIHALA/D_540_5493_644172.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 3, \"d_dod\": \"2012-02-13\", \"gender\": \"F\", \"age\": 65, \"d_deceased_nme_kn\": \"ರತ್ನವ್ವ ಮಾಲಾಪುರ\", \"d_father_kn\": \"ಮಲ್ಲಪ್ಪ ಮಾಲಾಪುರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಪ್ರಬಣ್ಣ ಮಾಲಾಪುರ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644172", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053563.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 71, \"b_dob\": \"1958-05-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಿದವ್ವ\", \"b_father_kn\": \"ಮುದುಕಪ್ಪ ಕರಿಯಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 72, \"b_dob\": \"1958-05-24\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರಾಮವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಮರದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 73, \"b_dob\": \"1958-05-25\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟಪ್ಪನಿಲಪ್ಪ\", \"b_father_kn\": \"ದುರ್ಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 74, \"b_dob\": \"1958-05-27\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಯಾಲವ್ವ\", \"b_father_kn\": \"ಚನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 75, \"b_dob\": \"1958-06-02\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸುಶೀಲ\", \"b_father_kn\": \"ಚನಬಸಪ್ಪ ಮಹಾಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 76, \"b_dob\": \"1958-06-04\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 77, \"b_dob\": \"1958-06-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚಂದಬಾರ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 78, \"b_dob\": \"1958-06-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕಾಳಪ್ಪ\", \"b_father_kn\": \"ವೀರಭದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 79, \"b_dob\": \"1958-06-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಶಿವಪುತ್ರಪ್ಪ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 80, \"b_dob\": \"1958-06-15\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕಸ್ತೂರಿ\", \"b_father_kn\": \"ನೀಲಕಂಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 10, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053563", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/DUMMIHALA/D_540_5493_644178.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 8, \"d_dod\": \"2012-11-19\", \"gender\": \"M\", \"age\": 65, \"d_deceased_nme_kn\": \"ಬಸಪ್ಪ ನಾಗಪ್ಪ ಸೋದ್ರಿ\", \"d_father_kn\": \"ನಾಗಪ್ಪ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಚಂದ್ರಪ್ಪ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644178", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053562.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 58, \"b_dob\": \"1958-05-03\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಿಂಗಪ್ಪ\", \"b_father_kn\": \"ಫಕೀರಪ್ಪ ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 61, \"b_dob\": \"1958-05-05\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಚಣವೆರಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 62, \"b_dob\": \"1958-05-05\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನೀಲವ್ವ\", \"b_father_kn\": \"ಸೀತಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 63, \"b_dob\": \"1958-05-09\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚನಮವ್ವ\", \"b_father_kn\": \"ಮಹ್ಮಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 64, \"b_dob\": \"1958-05-09\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸುರೇಶಪ್ಪ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 65, \"b_dob\": \"1958-05-09\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಎಳವ್ವ\", \"b_father_kn\": \"ಶಿದಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 66, \"b_dob\": \"1958-05-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಯಾಲಪ್ಪ ನೀಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 67, \"b_dob\": \"1958-05-18\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಾನಪ್ಪ\", \"b_father_kn\": \"ಕುಬರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 68, \"b_dob\": \"1958-05-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟವ್ವ\", \"b_father_kn\": \"ನಾಗಪ್ಪ ಗುಡಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 9, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053562", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644250.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 5, \"d_dod\": \"2012-02-15\", \"gender\": \"M\", \"age\": 80, \"d_deceased_nme_kn\": \"ಮಾನಪ್ಪ ಕಾಳಪ್ಪ ಬಡಿಗೇರ\", \"d_father_kn\": \"ಕಾಳಪ್ಪ ಬಡಿಗೇರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ರೂಪಾ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644250", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009357.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 12, \"b_dob\": \"2004-02-07\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಅಲ್ಲಮಾಸ\", \"b_father_kn\": \"ಅಮೀರ ಅಮ್ಮಾಜ ನಾಯ್ಕ\", \"mother_namekn\": \"ಹಸೀನಾ ಬಾನು\", \"info_name_kn\": \"ಶಹತಾಜ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009357", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644241.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 14, \"d_dod\": \"2012-07-18\", \"gender\": \"M\", \"age\": 26, \"d_deceased_nme_kn\": \"ರಮೇಶ ನಾಗಪ್ಪ ಹಲಗೇರಿ\", \"d_father_kn\": \"ನಾಗಪ್ಪ ಹಲಗೇರಿ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ತಿರಕಪ್ಪ ನಾಗಪ್ಪ ಮುಳಗುಂದ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644241", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009369.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 24, \"b_dob\": \"2004-02-22\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹ್ಮದ್ ಗೌಸಾ\", \"b_father_kn\": \"ಹಯಾತ್ ಸಾಬ್ ಶೇಖಾ\", \"mother_namekn\": \"ಶಹನಜ ಬಾನು\", \"info_name_kn\": \"ಶಹತಾಜ್ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009369", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009360.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 15, \"b_dob\": \"2004-02-08\", \"gender\": \"M\", \"b_baby_nme_kn\": null, \"b_father_kn\": \"ಹಜರತಲಿ\", \"mother_namekn\": \"ಶಬೀನಾ ಬಾನು\", \"info_name_kn\": \"ಶಹತಾಜ್ ಎನ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009360", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/KENGONDA/B_540_5493_0061667.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1958-01-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಸೋಮಶೇಖರ\", \"b_father_kn\": \"ಚನ್ನವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1958-01-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮುತ್ತಪ್ಪ\", \"b_father_kn\": \"ದೇವಿಂದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 3, \"b_dob\": \"1958-01-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಈರಪ್ಪ\", \"b_father_kn\": \"ಚನ್ನವೀರಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 4, \"b_dob\": \"1958-01-23\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ದುರ್ಗವ್ವ\", \"b_father_kn\": \"ಈಶ್ವರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 5, \"b_dob\": \"1958-01-23\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರಾಮಣ್ಣ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ ಭೀಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 5, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_0061667", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047270.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 12, \"b_dob\": \"1958-04-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮರಿಯವ್ವ\", \"b_father_kn\": null, \"mother_namekn\": \"ಕರಿಯವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 13, \"b_dob\": \"1958-04-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕಮಲವ್ವ\", \"b_father_kn\": \"ಬಸವಣ್ಣಪ್ಪ ಸಣ್ಣಬಸಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 14, \"b_dob\": \"1958-04-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಹದೇವಪ್ಪ\", \"b_father_kn\": \"ಮಲ್ಲಜ್ಜಪ್ಪ ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 15, \"b_dob\": \"1958-04-16\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ರುದ್ರಪ್ಪ\", \"b_father_kn\": \"ಸಿದ್ದಲಿಂಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 16, \"b_dob\": \"1958-04-14\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಭರಮಪ್ಪ\", \"b_father_kn\": \"ಸಿದ್ದಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 17, \"b_dob\": \"1958-05-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಶೆಂಕ್ರಪ್ಪ ಪುಟ್ಟಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 18, \"b_dob\": \"1958-05-09\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಭರಮವ್ವ\", \"b_father_kn\": \"ಶೆಂಕ್ರಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047270", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644242.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 13, \"d_dod\": \"2012-06-11\", \"gender\": \"F\", \"age\": 60, \"d_deceased_nme_kn\": \"ಹನುಮವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಶಂಕ್ರಪ್ಪ ಯತ್ತಿನಮನಿ\", \"mother_namekn\": null, \"info_name_kn\": \"ತಿರಕಪ್ಪ ನಾಗಪ್ಪ ಮುಳಗುಂದ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644242", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/DUMMIHALA/D_540_5493_644176.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 7, \"d_dod\": \"2012-10-02\", \"gender\": \"M\", \"age\": 31, \"d_deceased_nme_kn\": \"ರವಿ ಅಂಗಾರಗಟ್ಟಿ\", \"d_father_kn\": \"ಯಲ್ಲಪ್ಪ\", \"d_h_w_name_kn\": null, \"mother_namekn\": \"ಮಲ್ಲವ್ವ\", \"info_name_kn\": \"ಚಂದ್ರಪ್ಪ\", \"info_details\": null}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644176", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048560.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 15, \"b_dob\": \"1970-06-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಲಕ್ಷಮವ್ವ\", \"b_father_kn\": \"ಪರಮಪ್ಪ ನೀಲಪ್ಪ ಕನವಳ್ಳಿ\", \"mother_namekn\": \"ಎಲ್ಲವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048560", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047273.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 30, \"b_dob\": \"1958-10-29\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಮಲ್ಲಪ್ಪ\", \"b_father_kn\": \"ಸಿದ್ದಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 31, \"b_dob\": \"1958-10-30\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹನುಮವ್ವ\", \"b_father_kn\": \"ಬಸಪ್ಪ ಪರಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 32, \"b_dob\": \"1958-11-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಬಸವಂತಪ್ಪ ಗುರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 33, \"b_dob\": \"1958-11-04\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವರಾಜು\", \"b_father_kn\": \"ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 34, \"b_dob\": \"1958-11-08\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಲ್ಲವ್ವ\", \"b_father_kn\": \"ಹರಿಚಂದ್ರಪ್ಪ ಭರಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 36, \"b_dob\": \"1958-11-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ತಿರಕಪ್ಪ\", \"b_father_kn\": \"ನಂಜಪ್ಪ ಹಾಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 37, \"b_dob\": \"1958-11-11\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಗಂಗಿಮಾಳವ್ವ\", \"b_father_kn\": \"ಗೋಣೆಪ್ಪ\", \"mother_namekn\": \"ಹಿರಿಯವ್ವ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 38, \"b_dob\": \"1958-05-02\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಬಸವಣ್ಣಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 8, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047273", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_048020.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 25, \"b_dob\": \"1900-01-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಿಂಗನಗೌಡ\", \"b_father_kn\": \"ಮಲ್ಲನಗೌಡ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 26, \"b_dob\": \"1958-08-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗೋವಿಂದಪ್ಪ\", \"b_father_kn\": \"ಮಲ್ಲಪ್ಪ ಫಕೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 27, \"b_dob\": \"1958-08-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ದುಂಡೆಪ್ಪ\", \"b_father_kn\": \"ಚಂದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 28, \"b_dob\": \"1958-08-15\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಾಗಪ್ಪ\", \"b_father_kn\": \"ಭೀಮಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 29, \"b_dob\": \"1958-08-17\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಾವಂತ್ರವ್ವ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 30, \"b_dob\": \"1958-08-22\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹಿರಿಯವ್ವ\", \"b_father_kn\": \"ಭರಮಪ್ಪ ದುರಗಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 31, \"b_dob\": \"1958-06-04\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಲಲಿತವ್ವ\", \"b_father_kn\": \"ಸಣ್ಣಮೂಕಪ್ಪ ಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 7, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048020", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/AANUR/B_540_5493_047271.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 19, \"b_dob\": \"1958-06-26\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕಮಲವ್ವ\", \"b_father_kn\": \"ಹಾಲಪ್ಪ ಸಣ್ಣಮಲ್ಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 20, \"b_dob\": \"1958-07-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬೋಜಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಶೆಂಕ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 21, \"b_dob\": \"1958-07-26\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ನಾಗವ್ವ\", \"b_father_kn\": \"ಉಜ್ಜಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 23, \"b_dob\": \"1958-08-14\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಕಮಲವ್ವ\", \"b_father_kn\": \"ಉಜ್ಜಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 22, \"b_dob\": \"1958-08-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಪುಟ್ಟವ್ವ\", \"b_father_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 5, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_047271", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH_HAVERI_A3_AGASANAHALLI/AGASANAHALLI/B_540_5493_048555.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 1, \"b_dob\": \"1970-01-12\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ನಿಂಗಪ್ಪ\", \"b_father_kn\": \"ಪುಟ್ಟಪ್ಪ ಭರಮಪ್ಪ ಚೌಡಪ್ಪನವರ\", \"mother_namekn\": \"ಶಿವಕ್ಕ\", \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 2, \"b_dob\": \"1970-01-26\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಮಾಳವ್ವ\", \"b_father_kn\": \"ಗುಡ್ಡಪ್ಪ ಮಲ್ಲಪ್ಪ ಕನವಳ್ಳಿ\", \"mother_namekn\": \"ದುರಗವ್ವ\", \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1970\", \"total_records\": 2, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_048555", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MOTEBENNURU/B_540_5493_053560.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 42, \"b_dob\": \"1958-01-01\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚನಬಸಪ್ಪ\", \"b_father_kn\": null, \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 43, \"b_dob\": \"1958-04-05\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಕುಂತಲಾ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ ಶಿವಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 44, \"b_dob\": \"1958-04-05\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಚನ್ನಬಸಪ್ಪ\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ ಚನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 45, \"b_dob\": \"1958-04-07\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ಹನುಮಪ್ಪ ಹುಚ್ಚಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 46, \"b_dob\": \"1958-04-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವರಾಜು\", \"b_father_kn\": \"ರುದ್ರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 47, \"b_dob\": \"1958-04-10\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ರತ್ನವ್ವ\", \"b_father_kn\": \"ಯಾಲಪ್ಪ ರಾಮಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 48, \"b_dob\": \"1958-04-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪರಕೇಶ\", \"b_father_kn\": \"ಯಲ್ಲಪ್ಪ ಹನುಮಜಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 49, \"b_dob\": \"1958-04-11\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಗುಡಪ್ಪ\", \"b_father_kn\": \"ನಿಂಗಪ್ಪ ಹಾಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 50, \"b_dob\": \"1958-04-12\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಶಾಂತವ್ವ\", \"b_father_kn\": \"ಮಾಳಪ್ಪ ಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 9, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_053560", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/CHINNIKATTI/D_540_5493_644251.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 4, \"d_dod\": \"2012-02-08\", \"gender\": \"F\", \"age\": 58, \"d_deceased_nme_kn\": \"ಗಂಗವ್ವ\", \"d_father_kn\": null, \"d_h_w_name_kn\": \"ಗುಡ್ಡಪ್ಪ ನಿಂಗಪ್ಪ ಬಾಬಾ ಪುರ\", \"mother_namekn\": null, \"info_name_kn\": \"ಭರಮಪ್ಪ ಗುಡ್ಡಪ್ಪ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644251", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. d_dod - Date of death. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. age - Age of the deceased person at the time of death (a number)\n5. d_deceased_nme_kn - Deceased person's name ONLY in Kannada script. Do NOT include father's name or any other text.\n6. d_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include caste, village, or other details.\n7. d_h_w_name_kn - Husband's or Wife's name ONLY in Kannada script. Extract ONLY the person's name. This field may be empty for children.\n8. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n9. info_name_kn - Informant's name ONLY in Kannada. The person who reported the death. Extract ONLY the name.\n10. info_details - Informant's relationship or address in Kannada.\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields: extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"d_dod\": \"1958-03-17\",\n \"gender\": \"M\",\n \"age\": 65,\n \"d_deceased_nme_kn\": \"ಹೆಸರು\",\n \"d_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"d_h_w_name_kn\": \"ಗಂಡ/ಹೆಂಡತಿ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/DEATH_A5_HAVERI_KAGINELE HOBLI_SAMPLE_TEST/KAGINELE HOBLI/KALAGONDA/D_540_5493_644258.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 4, \"d_dod\": \"2012-12-03\", \"gender\": \"M\", \"age\": 60, \"d_deceased_nme_kn\": \"ಬಸಪ್ಪ ನೀಲಪ್ಪ\", \"d_father_kn\": \"ನೀಲಪ್ಪ ತಳವಾರ\", \"d_h_w_name_kn\": null, \"mother_namekn\": null, \"info_name_kn\": \"ಎನ್ ಎಸ್ ಹಟಮಣಿ\", \"info_details\": \"Chinnikatti\"}], \"page_info\": {\"year\": \"2012\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "D_D_540_5493_644258", "record_type": "D"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/BIRTH AND DEATH_A3_BYADAGI HOBLI/BYADAGI HOBLI/MALLAURU/B_540_5493_060781.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 86, \"b_dob\": \"1958-12-01\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಭೂಮಿಮ\", \"b_father_kn\": \"ಜಮಲಸಾಬ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 87, \"b_dob\": \"1958-12-06\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಹತನತುಬಿ\", \"b_father_kn\": \"ಗಸುನಾಬಿ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 88, \"b_dob\": \"1958-12-10\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಕೊತ್ರಯ್ಯ\", \"b_father_kn\": \"ಚನ್ನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 89, \"b_dob\": \"1958-12-16\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಚಂದ್ರವ್ವ\", \"b_father_kn\": \"ಚನ್ನವೀರಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 90, \"b_dob\": \"1958-12-26\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಉಲಸಪ್ಪ\", \"b_father_kn\": \"ಚನ್ನಬಸಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 91, \"b_dob\": \"1958-12-29\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಬಸವಂತಪ್ಪ\", \"b_father_kn\": \"ಹನುಮಂತಯ್ಯ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 92, \"b_dob\": \"1958-12-29\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಪುಟ್ಟಪ್ಪ\", \"b_father_kn\": \"ನೀಲಪ್ಪ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}, {\"reg_no\": 93, \"b_dob\": \"1958-12-29\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸುಂದ್ರಾಬಾಯಿ\", \"b_father_kn\": \"ಧ್ಯಮಣ್ಣ\", \"mother_namekn\": null, \"info_name_kn\": null, \"info_details\": null}], \"page_info\": {\"year\": \"1958\", \"total_records\": 8, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_060781", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009366.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 21, \"b_dob\": \"2004-02-15\", \"gender\": \"F\", \"b_baby_nme_kn\": \"ಸಬೆಲಾ ಬಾನು\", \"b_father_kn\": \"ಮಹ್ಮದ್ ಯಾಸಿನಾ ಮಹ್ಮದ್ ಅಲ್ಲಾ ವೇಲೂರಾ\", \"mother_namekn\": \"ಸೀಮಾ ಕೌಸರ\", \"info_name_kn\": \"ಮಹಮದ್ ಇಬ್ರಾಹಿಂ ಬಾಬು ಸಾಬ\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009366", "record_type": "B"}
|
| {"messages": [{"role": "user", "content": [{"type": "text", "text": "You are an expert OCR system specialized in reading handwritten Kannada text from historical birth/death register documents.\n\nPRIMARY GOAL: faithfully read the KANNADA handwriting.\n- Preserve Kannada script exactly as written.\n- Do NOT transliterate Kannada into English.\n- Do NOT modernize spelling or correct names.\n- If a field cannot be read confidently, output null.\n- Never hallucinate missing text.\n- Keep every physical register row separate.\n\nAnalyze this image and extract ALL records (rows) visible, reading from TOP to BOTTOM.\n\nFor each record/row, extract ONLY these fields:\n1. reg_no - Registration number (the serial number in the first column, a number)\n2. b_dob - Date of birth. Convert to YYYY-MM-DD format. Dates on the register are typically DD.MM.YYYY or DD/MM/YYYY.\n3. gender - M for Male (ಗಂಡು / ಗಂ) or F for Female (ಹೆಣ್ಣು / ಹೆ). Look carefully at the gender column or abbreviation.\n4. b_baby_nme_kn - Baby's name ONLY in Kannada script. Do NOT include father's name or any other text here.\n5. b_father_kn - Father's name ONLY in Kannada script. Extract ONLY the person's name. Do NOT include grandfather's name, caste, village, occupation, or any other details.\n6. mother_namekn - Mother's name ONLY in Kannada script. Extract ONLY the person's name.\n7. info_name_kn - Informant's name ONLY in Kannada. The person who reported the birth. Extract ONLY the name.\n8. info_details - Informant's relationship or address in Kannada (e.g. ತಂದೆ, ತಾಯಿ, etc.)\n\nCRITICAL RULES:\n- Each field should contain ONLY what belongs in that field, nothing extra.\n- For name fields (b_baby_nme_kn, b_father_kn, mother_namekn, info_name_kn): extract ONLY the person's name. Do NOT add caste, village, lineage, or address.\n- If a field is empty or illegible in the register, use null.\n- Multiple records may be on one page - extract ALL of them in order from top to bottom.\n\nReturn the data as a JSON object with this exact structure:\n{\n \"records\": [\n {\n \"reg_no\": 1,\n \"b_dob\": \"1958-01-04\",\n \"gender\": \"M\",\n \"b_baby_nme_kn\": \"ಹೆಸರು\",\n \"b_father_kn\": \"ತಂದೆ ಹೆಸರು\",\n \"mother_namekn\": \"ತಾಯಿ ಹೆಸರು\",\n \"info_name_kn\": \"ಮಾಹಿತಿದಾರ\",\n \"info_details\": \"ವಿವರಗಳು\"\n }\n ],\n \"page_info\": {\n \"year\": \"1958\",\n \"total_records\": 5,\n \"notes\": \"Any observations about the page quality or issues\"\n }\n}\n\nReturn ONLY valid JSON, no other text."}, {"type": "image", "image": "images/Sample Data for kannada OCR/Birth_A5_Haveri_KAGINELE_Sample Test/B_540_5493_009372.jpg"}]}, {"role": "assistant", "content": [{"type": "text", "text": "{\"records\": [{\"reg_no\": 27, \"b_dob\": \"2004-02-23\", \"gender\": \"M\", \"b_baby_nme_kn\": \"ಇಮಾಮ ಹುಸೇನ\", \"b_father_kn\": \"ಅನ್ವರ್ ಸಾಬ್ ಅತ್ತಾರ್\", \"mother_namekn\": \"ಫಾಥೋಬಿ\", \"info_name_kn\": \"ಶೇಕಮುಕ್ತ ಮಹ್ಮದ್ ಗೌಸಾ ಅತ್ತಾರ್\", \"info_details\": \"Kaginelli\"}], \"page_info\": {\"year\": \"2004\", \"total_records\": 1, \"notes\": \"Ground truth from source Excel\"}}"}]}], "id": "B_B_540_5493_009372", "record_type": "B"}
|
|
|
