Recruitment Requirement Extraction Evaluation Report
1. Executive Summary
- Dataset Evaluated: 25 Mixed-Format Placement Announcements (.txt, .eml, .pdf, .docx)
- Announcements Fully Extracted (Zero Manual Entry): 25 / 25 (100.0%)
- Target Requirement: At least 20 of 25 announcements without manual re-typing (Criterion 3).
- Target Status: PASSED (Achieved 25/25 fully offline).
- Total Benchmark Execution Time: 0.366 seconds (14.6 ms/document).
- Cloud/Network Calls: 0 (100% offline regex & semantic dictionary architecture).
- LLM Dependency: None (LLM Adapter OFF by default; target strictly achieved deterministically).
2. Per-Field Precision and Recall
| Field Name | Evaluated | Correct (TP) | False Pos | False Neg | Precision | Recall | F1 Score |
|---|---|---|---|---|---|---|---|
company_name |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
drive_date |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
mode |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
ctc_lpa |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
cgpa_min |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
tenth_percentage_min |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
twelfth_percentage_min |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
active_backlogs_max |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
historical_backlogs_max |
8 | 8 | 0 | 0 | 100.0% | 100.0% | 100.0% |
allowed_branches |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
gap_years_max |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
rounds_count |
25 | 25 | 0 | 0 | 100.0% | 100.0% | 100.0% |
3. Document-by-Document Breakdown
| Filename | Company | Extracted Status | Confidence | Ambiguities Flagged | Review Items |
|---|---|---|---|---|---|
01_google_announcement.txt |
Full Extract (Zero Manual Entry) | 0.96 | 0 | 0 | |
02_microsoft_announcement.txt |
Microsoft | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
03_amazon_announcement.txt |
Amazon | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
04_goldman_sachs_announcement.txt |
Goldman Sachs | Full Extract (Zero Manual Entry) | 0.96 | 0 | 0 |
05_oracle_announcement.txt |
Oracle | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
06_qualcomm_announcement.txt |
Qualcomm | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
07_cisco_announcement.txt |
Cisco | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
08_deloitte_usi_email.eml |
Deloitte USI | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
09_tata_consultancy_services_email.eml |
Tata Consultancy Services | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
10_infosys_power_programmer_email.eml |
Infosys Power Programmer | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
11_wipro_turbo_email.eml |
Wipro Turbo | Full Extract (Zero Manual Entry) | 0.95 | 1 | 1 |
12_accenture_advanced_tech_email.eml |
Accenture Advanced Tech | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
13_siemens_email.eml |
Siemens | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
14_larsen_and_toubro_spec.pdf |
Larsen & Toubro | Full Extract (Zero Manual Entry) | 0.96 | 0 | 0 |
15_texas_instruments_spec.pdf |
Texas Instruments | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
16_intel_spec.pdf |
Intel | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
17_morgan_stanley_spec.pdf |
Morgan Stanley | Full Extract (Zero Manual Entry) | 0.96 | 0 | 0 |
18_samsung_randd_spec.pdf |
Samsung R&D | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
19_tata_motors_spec.pdf |
Tata Motors | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
20_reliance_industries_jd.docx |
Reliance Industries | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
21_itc_limited_jd.docx |
ITC Limited | Full Extract (Zero Manual Entry) | 0.96 | 0 | 0 |
22_biocon_jd.docx |
Biocon | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
23_zs_associates_jd.docx |
ZS Associates | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
24_barclays_jd.docx |
Barclays | Full Extract (Zero Manual Entry) | 0.95 | 0 | 0 |
25_atlassian_jd.docx |
Atlassian | Full Extract (Zero Manual Entry) | 0.96 | 0 | 0 |
4. Ambiguity and Review Queue Governance
In compliance with institutional placement standards:
- Never Guess Silently: Phrasings such as 'throughout' or 'at time of joining' are explicitly identified as ambiguities and automatically appended to the review queue with suggested clarification actions.
- Side-by-Side Review Contract: The officer UI receives the verbatim sentence provenance alongside every parsed field, enabling 1-click accept, edit, or reject.
- Feedback Audit Log: Every correction or override is persisted in the audit trail with officer ID and timestamp.
5. Offline & Privacy Assurance
- All text parsing, document decompression, and table extractions execute locally.
- No student data, corporate specifications, or salary details leave the local host environment.