Spaces:
Runtime error
Runtime error
File size: 59,060 Bytes
17847d4 62b327e 17847d4 62b327e 17847d4 f1515e9 17847d4 62b327e 17847d4 f1348a5 17847d4 8e6bdfb 17847d4 8e6bdfb 17847d4 8e6bdfb 17847d4 8e6bdfb 17847d4 8e6bdfb 17847d4 62b327e 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 b40f51b 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 62b327e 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 62b327e f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 62b327e f1348a5 17847d4 f1348a5 17847d4 f1348a5 5078721 f1348a5 5078721 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 f1348a5 17847d4 ce0eb39 17847d4 ce0eb39 17847d4 ce0eb39 17847d4 ce0eb39 17847d4 ce0eb39 17847d4 f1348a5 5078721 f1348a5 5757067 f1348a5 5078721 f1348a5 5757067 f1348a5 5757067 f1348a5 5078721 f1348a5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 | """
DSPy-based Form Generation System
==================================
This module uses DSPy to generate forms from natural language queries.
It includes a multi-stage pipeline:
1. Planner: Converts user query to form plan with components
2. Component Signature: Generates renderable form components with metadata
3. Conditional Logic: Defines relationships between components
FRONTEND TECH STACK:
- React with TypeScript
- Headless UI (@headlessui/react) for accessible dialogs, modals, and interactive components
- React Markdown (react-markdown) with remark-gfm for markdown rendering
- Tally-inspired minimal design with inline styles (no CSS frameworks)
- Purple accent color (#9333ea) throughout
- Clean, generous whitespace (48px between questions)
- Subtle borders and transitions
"""
import dspy
import json
import logging
import os
import warnings
from typing import Dict, Any, List, Optional
# Suppress Pydantic serialization warnings from DSPy/LiteLLM
warnings.filterwarnings("ignore", message="Pydantic serializer warnings")
warnings.filterwarnings("ignore", category=UserWarning, module="pydantic")
# Set up logger for the module
logger = logging.getLogger("dspy_forms")
logger.setLevel(logging.WARNING)
if not logger.handlers:
ch = logging.StreamHandler()
formatter = logging.Formatter('[%(asctime)s][%(levelname)s] %(message)s')
ch.setFormatter(formatter)
logger.addHandler(ch)
# ============================================================================
# FORM GENERATION SIGNATURES
# ============================================================================
class FormPlannerSignature(dspy.Signature):
"""
You are an AI form designer that creates comprehensive form plans from user descriptions.
TASK:
Convert a natural language form request into a structured JSON form plan with component specifications.
OUTPUT FORMAT (JSON):
{
"title": "Form Title",
"description": "Brief description of form purpose",
"submit_button_text": "Submit" or custom text,
"components": [
{
"component_id": "comp_1",
"type": "question_type",
"brief": "Brief description of what this component captures",
"required": true/false,
"order": 0
}
],
"conditional_logic": [
{
"trigger_component_id": "comp_1",
"target_component_id": "comp_3",
"condition_type": "equals",
"condition_value": "Yes",
"action": "show"
}
]
}
COMPONENT TYPES (use exact strings):
- short_answer: Single line text
- long_answer: Multi-line text
- multiple_choice: Radio buttons (one selection)
- checkboxes: Multiple selections
- dropdown: Select dropdown
- multi_select: Multi-select dropdown
- number: Numeric input
- email: Email validation
- phone: Phone number
- link: URL input
- file_upload: File attachment
- date: Date picker
- time: Time picker
- linear_scale: Scale (1-10)
- matrix: Grid questions
- rating: Star rating
- payment: Payment field
- signature: Digital signature
- ranking: Rank items
- wallet_connect: Web3 wallet
- button: Custom action button
CONDITIONAL LOGIC RULES:
- condition_type: equals, not_equals, contains, not_contains, greater_than, less_than, is_empty, is_not_empty
- action: show, hide
CRITICAL GUIDELINES:
1. **FOLLOW USER'S EXACT SPECIFICATIONS** - If user says "Valima, Nikkah, Barat", use those EXACT terms, not substitutes
2. **DO NOT SUBSTITUTE** terms - Use the exact names, labels, and options the user provides
3. **USE DIVERSE QUESTION TYPES** - Mix short_answer, multiple_choice, checkboxes, dropdown, date, rating, etc.
4. **USE CHECKBOXES for multiple selections** - When users can select multiple items (e.g., "Which events will you attend?")
5. **USE MULTIPLE_CHOICE for single selection** - When users pick one option (e.g., "Are you attending? Yes/No")
6. Choose appropriate component types based on the data being collected
7. Mark essential components as required
8. Order components logically (general to specific)
9. **INCLUDE AT LEAST 5 COMPONENTS** - Forms should have a minimum of 5 questions/components. You can reuse the same component type multiple times if appropriate (e.g., multiple short_answer fields for different information)
10. Add conditional logic where it makes sense (e.g., show follow-up based on previous answer)
11. Generate component_ids like "comp_1", "comp_2", etc.
12. **PREFER STRUCTURED INPUTS** over text when possible (use checkboxes, dropdowns, ratings instead of open text)
13. **KEEP IT CONCISE** - Descriptions should be 1-2 sentences max. Option lists should have 3-8 items max, not exhaustive lists.
**IMPORTANT: PRESERVE USER'S TERMINOLOGY**
- If user mentions specific event names (Valima, Nikkah, Barat, etc.), use those EXACT names
- If user mentions specific options or choices, use those EXACT options
- Do NOT replace cultural, local, or specific terms with generic alternatives
- Do NOT assume Western/US conventions unless specifically requested
COMMON FORM PATTERNS (adapt to user's context):
- Contact forms: name (short_answer), email, phone, message (long_answer)
- Registration forms: name, email, password, interests (checkboxes), preferences (dropdown)
- Feedback forms: name (optional), rating, satisfaction (linear_scale), comments (long_answer)
- Applications: name, email, skills (checkboxes), experience level (dropdown)
- Event registration: name, email, events attending (checkboxes), meal preference (dropdown), dietary restrictions (checkboxes)
- Survey forms: demographics (dropdowns), agreement scales (linear_scale), multiple choice questions, open-ended responses (long_answer)
- RSVP forms: attendance (multiple_choice: Yes/No), guest count (number), dietary needs (checkboxes)
FORM TITLE BEST PRACTICES:
- **ALWAYS wrap title in double asterisks** for bold formatting: "**Your Form Title**"
- Be specific and clear (e.g., "**Customer Feedback Survey**" not "Form")
- Use action words when appropriate (e.g., "**Submit Your Application**", "**Register for Event**")
- Keep under 60 characters for readability
- Match the tone to the form purpose (professional for applications, friendly for feedback)
CRITICAL: The title field MUST always start and end with ** to render as bold markdown.
Example: "title": "**Wedding RSVP**" NOT "title": "Wedding RSVP"
Always include a meaningful form title and description that clearly explains the purpose.
"""
user_query = dspy.InputField(desc="Natural language description of the form needed")
form_plan = dspy.OutputField(desc="JSON form plan with title, description, components list, and conditional_logic")
class ComponentTypeDetectorSignature(dspy.Signature):
"""
Detect the correct component type from a user's add request.
COMPONENT TYPES (return EXACTLY one of these strings):
- short_answer: Single line text (name, title, brief response)
- long_answer: Multi-line text (description, feedback, detailed response)
- multiple_choice: Radio buttons - ONE selection only (yes/no, single option)
- checkboxes: Multiple selections allowed (select all that apply)
- dropdown: Single select from dropdown list
- multi_select: Multi-select dropdown
- number: Numeric input (age, quantity, amount)
- email: Email address input
- phone: Phone number input
- link: URL/website input
- file_upload: File attachment (resume, document, image)
- date: Date picker (birthday, event date)
- time: Time picker (appointment time)
- linear_scale: Scale/slider (1-5, 1-10 rating with labels)
- matrix: Grid questions (rows x columns)
- rating: Star rating (1-5 stars)
- payment: Payment field
- signature: Digital signature capture
- ranking: Rank/order items by preference
- wallet_connect: Web3 wallet connection
- button: Custom action button
DETECTION RULES:
1. "checkbox" or "checkboxes" or "select multiple" β checkboxes
2. "radio" or "single choice" or "choose one" β multiple_choice
3. "dropdown" or "select" (single) β dropdown
4. "rating" or "star" or "stars" β rating
5. "scale" or "slider" or "1 to 10" or "rate from" β linear_scale
6. "email" β email
7. "phone" or "telephone" or "mobile" β phone
8. "date" or "birthday" or "when" β date
9. "time" or "what time" β time
10. "file" or "upload" or "attachment" or "resume" or "document" β file_upload
11. "signature" or "sign" β signature
12. "rank" or "order" or "prioritize" β ranking
13. "matrix" or "grid" β matrix
14. "number" or "amount" or "quantity" or "age" or "how many" β number
15. "website" or "url" or "link" β link
16. "long" or "paragraph" or "describe" or "explain" or "feedback" β long_answer
17. Default for text questions β short_answer
"""
user_query = dspy.InputField(desc="User's add component request")
component_type = dspy.OutputField(desc="Exact component type string from the list above")
reasoning = dspy.OutputField(desc="Brief explanation of why this type was chosen")
class ComponentSignatureGenerator(dspy.Signature):
"""
Generate detailed component specifications that the frontend can render.
CRITICAL: Output ONLY valid JSON. NO comments (no // or /* */). NO trailing commas.
FRONTEND CAPABILITIES:
- Headless UI for modals and interactive components
- Markdown support in question text and descriptions
- Inline editing with QuestionEditor component
- Tally-inspired minimal design with clean typography
- Purple accent (#9333ea) for focused/selected states
INPUT:
- component_brief: Brief description from planner
- component_type: Type of component
- component_id: Unique identifier
- form_context: Overall form context
OUTPUT FORMAT (strict JSON, no comments):
{
"component_id": "comp_1",
"question_type": "file_upload",
"question_text": "Upload your resume/CV",
"description": "Accepted formats: PDF, DOC, DOCX (max 10MB)",
"required": true,
"settings": {
"file_types": [".pdf", ".doc", ".docx"],
"max_file_size": 10485760
}
}
SETTINGS BY TYPE (include ONLY relevant settings):
- multiple_choice/checkboxes/dropdown: "choices": ["Option A", "Option B"]
- number/linear_scale: "min_value": 1, "max_value": 10
- text inputs: "placeholder": "Enter text here"
- linear_scale: "scale_min_label": "Low", "scale_max_label": "High"
- matrix: "rows": ["Row 1"], "columns": ["Col 1"]
- file_upload: "file_types": [".pdf", ".doc"], "max_file_size": 5242880
- ranking: "ranking_items": ["Item 1", "Item 2"]
RULES:
1. **USE USER'S EXACT TERMINOLOGY** - If the brief mentions specific terms (e.g., "Valima", "Nikkah", "Barat"), use those EXACT terms
2. **DO NOT SUBSTITUTE** cultural, local, or specific terms with generic alternatives
3. **MATCH COMPONENT TYPE TO DATA** - For multiple selections, use checkboxes; for single selection, use multiple_choice or dropdown
4. Provide choices exactly as described by user (or 3-5 sensible options if not specified)
5. Use descriptive labels for scales (e.g., "Not at all" to "Extremely")
6. Set reasonable limits for numbers and file sizes
7. Include helpful placeholder text for text inputs
8. Add description for complex components (matrix, ranking, file uploads)
9. Include validation rules where appropriate
10. CRITICAL: rows, columns, and ranking_items MUST be arrays of strings, never integers
11. Question text and descriptions support markdown (use **bold**, *italic*, lists, etc.)
12. Keep text clear and conversational - the design is minimal and clean
13. **KEEP OPTIONS SHORT** - Maximum 8 choices for dropdowns/checkboxes. Do NOT generate exhaustive lists.
14. **DESCRIPTIONS BRIEF** - Keep descriptions to 1-2 short sentences max. No long explanations.
13. **GENERATE COMPLETE CHOICES** - Always provide actual choice options, not placeholders like "Option 1", "Option 2"
EXAMPLES OF PRESERVING USER TERMINOLOGY:
- User says "Valima, Nikkah, Barat" β choices: ["Valima", "Nikkah", "Barat"] (NOT "Reception, Ceremony, etc.")
- User says "yes/no" β choices: ["Yes", "No"] (NOT "Agree/Disagree" unless asked)
- User mentions specific names/terms β use them exactly as written
"""
component_brief = dspy.InputField(desc="Brief description from planner")
component_type = dspy.InputField(desc="Component type")
component_id = dspy.InputField(desc="Unique component identifier")
form_context = dspy.InputField(desc="Overall form context for better component generation")
component_spec = dspy.OutputField(desc="Complete JSON component specification ready for frontend rendering")
class FormEditorSignature(dspy.Signature):
"""
Edit an existing form based on user requests.
FRONTEND CAPABILITIES:
- Headless UI for dialogs and modals
- Markdown support in all text fields
- Inline editing with QuestionEditor component
- Tally-inspired minimal design
- Purple accent color throughout
INPUT:
- edit_request: Natural language edit instruction
- current_form: JSON with form structure INCLUDING component summary at top level
The current_form JSON includes a "component_summary" field like:
[{"index": 1, "id": 123, "type": "checkboxes", "text": "Which options..."}]
EDIT TYPES:
1. Add components: Insert new components at appropriate position
2. Remove components: Delete specified components
3. Modify components: Change text, type, or settings
4. Reorder components: Change component sequence
5. Update form metadata: Change title, description, settings
6. Update conditional logic: Add/remove/modify rules
HOW TO FIND COMPONENTS:
- Look at "component_summary" in current_form for quick reference
- Match user's request to the most relevant component by text similarity or type
- If user says "the checkbox" find the component with type "checkboxes"
- If user references by number (e.g., "question 2"), use the index in component_summary
- If ambiguous, apply edit to the MOST LIKELY matching component
COMPONENT TYPES (for reference):
- short_answer, long_answer, multiple_choice, checkboxes, dropdown
- multi_select, number, email, phone, link, file_upload
- date, time, linear_scale, matrix, rating, payment, signature, ranking
OUTPUT FORMAT (JSON):
{
"title": "Updated title",
"description": "Updated description",
"components": [
// Updated components array with component_id, type, settings, etc.
],
"conditional_logic": [
// Updated conditional logic rules
],
"submit_button_text": "Submit"
}
RULES:
1. Preserve existing components unless explicitly asked to change
2. Maintain component_id consistency
3. Update conditional logic if component_ids change
4. Apply changes precisely as requested
5. Return complete updated form structure
6. When changing component type, preserve the question_text and adapt settings appropriately
"""
edit_request = dspy.InputField(desc="User's edit instruction")
current_form = dspy.InputField(desc="Current form structure as JSON with component_summary for quick reference")
updated_form = dspy.OutputField(desc="Complete updated form structure as JSON")
changes_made = dspy.OutputField(desc="Summary of changes applied")
class QuestionEditorSignature(dspy.Signature):
"""
Edit/regenerate a single question component based on user's edit instructions.
This signature is specifically designed for editing individual questions/components.
It takes the current component state and user's edit prompt, then generates a new
component specification that may have a different type and/or updated metadata.
INPUT:
- current_question_type: The current question type (e.g., "short_answer", "multiple_choice", "checkboxes")
- current_question_text: Current question text/label
- current_description: Current description/helper text (if any)
- current_settings: Current component settings as JSON (choices, validation, etc.)
- current_required: Whether the question is currently required
- user_edit_prompt: User's instruction for how to edit the question (e.g., "Make this a long text", "Add more options", "Change to multiple choice")
- form_context: Context about the form (title, description, other questions)
OUTPUT FORMAT (JSON):
{
"question_type": "new_or_same_type", // Can be same or different type based on user's request
"question_text": "Updated question text (markdown supported)",
"description": "Updated description (markdown supported, optional)",
"required": true/false,
"settings": {
// Type-specific settings matching the NEW question_type
"choices": ["Option 1", "Option 2"], // For multiple_choice, checkboxes, dropdown
"min_value": 1, // For number, linear_scale
"max_value": 10,
"placeholder": "Enter text here", // For text inputs
"scale_min_label": "Not at all", // For linear_scale
"scale_max_label": "Extremely",
"rows": ["Row 1", "Row 2"], // For matrix (MUST be array of strings)
"columns": ["Col 1", "Col 2"], // For matrix (MUST be array of strings)
"file_types": [".pdf", ".doc"], // For file_upload
"max_file_size": 5242880, // bytes
"ranking_items": ["Item 1", "Item 2"] // For ranking (MUST be array of strings)
}
}
QUESTION TYPES (use exact strings):
- short_answer: Single line text
- long_answer: Multi-line text
- multiple_choice: Radio buttons (one selection)
- checkboxes: Multiple selections
- dropdown: Select dropdown
- multi_select: Multi-select dropdown
- number: Numeric input
- email: Email validation
- phone: Phone number
- link: URL input
- file_upload: File attachment
- date: Date picker
- time: Time picker
- linear_scale: Scale (1-10)
- matrix: Grid questions
- rating: Star rating
- payment: Payment field
- signature: Digital signature
- ranking: Rank items
- wallet_connect: Web3 wallet
- button: Custom action button
RULES:
1. **FOLLOW USER'S EDIT INSTRUCTIONS PRECISELY** - If user says "make this a long text", change type to "long_answer"
2. **PRESERVE EXISTING DATA WHEN APPROPRIATE** - If user doesn't specify changing type, keep the same type unless it makes sense to change
3. **UPDATE SETTINGS TO MATCH NEW TYPE** - If type changes, generate appropriate settings for the new type
4. **PRESERVE USER'S TERMINOLOGY** - Keep exact terms from current question unless user explicitly asks to change them
5. **GENERATE COMPLETE SETTINGS** - Always provide complete settings for the question type (e.g., all choices for multiple_choice)
6. **MAINTAIN CONSISTENCY** - Keep question text, description style consistent with form context
7. **SUPPORT MARKDOWN** - Question text and description support markdown (use **bold**, *italic*, etc.)
8. **CRITICAL: rows, columns, ranking_items MUST be arrays of strings**, never integers
9. If user asks to "add more options" or "add choices", expand the choices array
10. If user asks to change type, ensure settings match the new type requirements
EXAMPLES:
- User says "Make this a long text" β Change type to "long_answer", remove choices if present
- User says "Add more options" β Keep type, add more items to choices array
- User says "Change to multiple choice" β Change type to "multiple_choice", generate appropriate choices
- User says "Make it required" β Keep everything same, set required=true
"""
current_question_type = dspy.InputField(desc="Current question type")
current_question_text = dspy.InputField(desc="Current question text")
current_description = dspy.InputField(desc="Current description (can be empty)")
current_settings = dspy.InputField(desc="Current settings as JSON string")
current_required = dspy.InputField(desc="Whether question is currently required (true/false)")
user_edit_prompt = dspy.InputField(desc="User's instruction for editing the question")
form_context = dspy.InputField(desc="Form context (title, description, other questions)")
question_spec = dspy.OutputField(desc="Complete JSON question specification with potentially new type and metadata")
class ComponentMatcherSignature(dspy.Signature):
"""Match user query to a specific component in the form. Return component_id or null."""
query = dspy.InputField(desc="user query about a component")
components = dspy.InputField(desc="JSON: [{'id':'comp_1','type':'short_answer','text':'Question text'},...]")
component_id = dspy.OutputField(desc="Matching component_id or null if no match")
class FormChatRouterSignature(dspy.Signature):
"""Route user queries for form editing OR response analysis.
ROUTING RULES:
**FORM EDITING:**
- **add_component**: User wants to ADD a new form component/question/field.
Keywords: "add", "include", "insert", "create field", "add question"
Examples: "Add an email field", "Include a phone number question", "Add a rating field"
- **edit_component**: User wants to MODIFY an existing component.
Keywords: "change", "modify", "update", "edit", "make it", "adjust"
Examples: "Change the email field to required", "Update the rating scale to 1-5"
**RESPONSE ANALYSIS:**
- **analyze_responses**: User wants to analyze, summarize, or query form responses/submissions.
Keywords: "responses", "submissions", "answers", "who answered", "how many", "summarize", "filter", "average", "sentiment", "export"
Examples: "Summarize my responses", "How many people submitted?", "Show me responses where rating > 3", "What's the average rating?", "Export responses", "What did people say?", "Analyze the feedback"
**OTHER:**
- **general_query**: General questions about the form structure or capabilities.
Examples: "What fields are in this form?", "How does conditional logic work?"
- **need_more_clarity**: Query is ambiguous, unclear, or possibly irrelevant.
IMPORTANT: Carefully distinguish between form EDITING and response ANALYSIS.
"""
user_query = dspy.InputField(desc="The user's query about the form")
form_context = dspy.InputField(desc="Context about the form structure and response count")
query_type = dspy.OutputField(desc="One of: 'add_component', 'edit_component', 'analyze_responses', 'general_query', 'need_more_clarity'")
reasoning = dspy.OutputField(desc="Brief explanation of why this route was chosen")
class GeneralFormQASignature(dspy.Signature):
"""Answer general questions about the form briefly and concisely.
IMPORTANT RULES:
- Keep answers SHORT - maximum 2-3 sentences
- Be direct and to the point
- Do NOT provide long explanations or lists
- If asked about something unrelated to the form, briefly redirect to form editing
- NEVER generate long content like lists of items, detailed descriptions, or educational content
"""
user_query = dspy.InputField(desc="The user's question (max 1000 chars)")
form_context = dspy.InputField(desc="Current form structure")
answer = dspy.OutputField(desc="Brief answer in 2-3 sentences MAX. Keep under 500 characters.")
CLARITY_RESPONSE = (
"I'm sorry, I couldn't understand your request.\n\n"
"#### Here's what I can help you with:\n"
"- **Add** a new component or field to your form\n"
"- **Edit** or **modify** an existing component\n"
"- **Analyze responses** - summarize, filter, or query submissions\n"
"- Answer **general questions** about your form\n\n"
"Please clarify what you'd like to do!"
)
NO_RESPONSES_MESSAGE = (
"No responses yet! Once people submit your form, I can help you:\n\n"
"- **Summarize** all responses\n"
"- **Filter** responses by criteria\n"
"- **Calculate** statistics (averages, counts, etc.)\n"
"- **Analyze sentiment** of text answers\n"
"- **Export** custom data sets\n\n"
"Share your form to start collecting responses!"
)
class FormChatFunction(dspy.Module):
"""
Unified chat module for form editing AND response analysis.
Uses GPT-4o-mini for:
1. Routing queries to appropriate handler
2. Summarizing results
3. Deciding whether to use memory
Condition: If response_count == 0, analyze_responses route returns NO_RESPONSES_MESSAGE.
"""
def __init__(self):
# Form editing signatures
self.type_detector = dspy.Predict(ComponentTypeDetectorSignature)
self.add_component = dspy.Predict(ComponentSignatureGenerator)
self.edit_form = dspy.Predict(FormEditorSignature)
self.general_qa = dspy.Predict(GeneralFormQASignature)
# Router
self.router = dspy.Predict(FormChatRouterSignature)
# Response analysis signatures
self.sql_generator = dspy.Predict(SQLGeneratorSignature)
self.summarizer = dspy.Predict(ResponseSummarizerSignature)
self.CLARITY_RESPONSE = CLARITY_RESPONSE
self.NO_RESPONSES_MESSAGE = NO_RESPONSES_MESSAGE
async def aforward(
self,
user_query: str,
form_context: str,
current_form: dict = None,
response_count: int = 0,
chat_history: List[Dict] = None,
response_data: dict = None
) -> dict:
"""
Unified chat handler for form editing and response analysis.
Args:
user_query: User's natural language request
form_context: JSON string of current form structure
current_form: Dict of current form structure
response_count: Number of responses (0 = no analysis)
chat_history: Previous messages for memory
response_data: Pre-loaded response data (schema, summary, duckdb conn)
Returns:
dict with route, response, and metadata
"""
user_query = user_query[:1000] if len(user_query) > 1000 else user_query
# Build enhanced context with response count
try:
context_dict = json.loads(form_context) if isinstance(form_context, str) else (form_context or {})
except:
context_dict = {}
context_dict['response_count'] = response_count
context_dict['has_responses'] = response_count > 0
enhanced_context = json.dumps(context_dict)
# === STEP 1: Route using GPT-4o-mini ===
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=500, temperature=0.3)):
route_result = self.router(user_query=user_query, form_context=enhanced_context)
query_type = route_result.query_type.lower().strip()
# Recheck if unclear
if 'clarity' in query_type or 'unclear' in query_type:
route_result = self.router(user_query=user_query, form_context=enhanced_context)
query_type = route_result.query_type.lower().strip()
result = {
'route': query_type,
'reasoning': getattr(route_result, 'reasoning', ''),
'response': None,
'data': None,
'use_memory': False,
'requires_response_analysis': False
}
# === STEP 2: Handle based on route ===
# --- RESPONSE ANALYSIS ---
if 'analyze' in query_type or 'response' in query_type:
# CONDITION: No responses = no analysis
if response_count == 0:
result['response'] = self.NO_RESPONSES_MESSAGE
result['route'] = 'no_responses'
return result
result['route'] = 'analyze_responses'
result['use_memory'] = True
result['requires_response_analysis'] = True
# VALIDATION: Check if DuckDB data is ready
if not response_data:
result['response'] = "Unable to analyze responses - data not loaded. Please try again."
result['duckdb_ready'] = False
return result
if not response_data.get('schema_description'):
result['response'] = "Unable to analyze responses - schema not available. Please try again."
result['duckdb_ready'] = False
return result
if not response_data.get('conn'):
result['response'] = "Unable to analyze responses - database connection not ready. Please try again."
result['duckdb_ready'] = False
return result
result['duckdb_ready'] = True
# Response data validated, proceed with analysis
schema = response_data['schema_description']
summary = response_data.get('summary', {})
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=1300)):
# Generate SQL
sql_result = self.sql_generator(
user_query=user_query,
schema_description=schema
)
result['sql_query'] = sql_result.sql_query
# Summarize
summary_result = self.summarizer(
query_results=json.dumps(summary, default=str),
user_query=user_query,
schema_description=schema
)
result['response'] = summary_result.summary
return result
# --- ADD COMPONENT ---
if 'add' in query_type:
result['route'] = 'add_component'
existing_ids = []
if current_form and 'questions' in current_form:
existing_ids = [f"comp_{i+1}" for i in range(len(current_form['questions']))]
new_id = f"comp_{len(existing_ids) + 1}"
# First, detect the component type from user query
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=200)):
type_result = self.type_detector(user_query=user_query)
detected_type = getattr(type_result, 'component_type', 'short_answer')
# Clean up type - ensure it's a valid type string
valid_types = [
'short_answer', 'long_answer', 'multiple_choice', 'checkboxes',
'dropdown', 'multi_select', 'number', 'email', 'phone', 'link',
'file_upload', 'date', 'time', 'linear_scale', 'matrix', 'rating',
'payment', 'signature', 'ranking', 'wallet_connect', 'button'
]
detected_type = detected_type.lower().strip().replace(' ', '_')
if detected_type not in valid_types:
detected_type = 'short_answer'
logger.info(f"Detected component type: {detected_type} for query: {user_query}")
add_context = json.dumps({
**context_dict,
'ORIGINAL_USER_REQUEST': user_query,
'DETECTED_COMPONENT_TYPE': detected_type,
'IMPORTANT': 'Use EXACT terms from user. Do NOT substitute cultural terms.'
})
# Now generate the component with the detected type
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=1000)):
component = self.add_component(
component_brief=f"{user_query}. Use exact terms from user. Component type MUST be: {detected_type}",
component_type=detected_type,
component_id=new_id,
form_context=add_context
)
result['response'] = component
return result
# --- EDIT COMPONENT ---
if 'edit' in query_type:
result['route'] = 'edit_component'
if not current_form:
result['response'] = "No form available to edit."
return result
# Build component summary and embed it into the form for easy matching
component_summary = []
if 'questions' in current_form:
for idx, q in enumerate(current_form['questions']):
component_summary.append({
'index': idx + 1,
'id': q.get('id', f'comp_{idx+1}'),
'type': q.get('question_type', 'unknown'),
'text': q.get('question_text', '')[:80], # Truncate for brevity
'required': q.get('required', False)
})
# Create form with embedded component summary
form_with_summary = {
'component_summary': component_summary,
**current_form
}
logger.info(f"Editing form with {len(component_summary)} components")
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=1500)):
edit_result = self.edit_form(
edit_request=user_query,
current_form=json.dumps(form_with_summary)
)
result['response'] = edit_result
return result
# --- GENERAL QUERY ---
if 'general' in query_type:
result['route'] = 'general_query'
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=500)):
qa_result = self.general_qa(
user_query=user_query,
form_context=form_context
)
answer = getattr(qa_result, 'answer', str(qa_result))
result['response'] = answer[:1000] if len(answer) > 1000 else answer
return result
# --- UNCLEAR ---
result['route'] = 'need_clarity'
result['response'] = self.CLARITY_RESPONSE
return result
def summarize_query_results(
self,
query_results: List[Dict],
user_query: str,
schema_description: str
) -> str:
"""Summarize SQL query results in natural language."""
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=800)):
result = self.summarizer(
query_results=json.dumps(query_results[:50], default=str),
user_query=user_query,
schema_description=schema_description
)
return result.summary
# ============================================================================
# FORM FIXER SIGNATURE
# ============================================================================
class FormFixerSignature(dspy.Signature):
"""
Fix validation errors in a generated form plan.
You are given a form that failed validation along with the specific errors.
Your job is to fix ONLY the issues mentioned in the validation errors while
preserving all other valid parts of the form.
VALID QUESTION TYPES (use ONLY these exact strings):
- short_answer: Single line text input
- long_answer: Multi-line text input
- multiple_choice: Radio buttons (one selection)
- checkboxes: Multiple selections allowed
- dropdown: Single select dropdown
- multi_select: Multi-select dropdown
- number: Numeric input
- email: Email address input
- phone: Phone number input
- link: URL input
- file_upload: File attachment
- date: Date picker
- time: Time picker
- linear_scale: Scale rating (1-10)
- matrix: Grid questions
- rating: Star rating
- payment: Payment field
- signature: Digital signature
- ranking: Rank items
- wallet_connect: Web3 wallet
- button: Action button
COMMON FIXES:
- "text" β "short_answer"
- "textarea" β "long_answer"
- "radio" β "multiple_choice"
- "checkbox" β "checkboxes"
- "select" β "dropdown"
- Missing question_type β add appropriate type based on question_text
- Missing question_text β generate appropriate text
- Missing component_id β generate as "comp_N"
RULES:
1. Fix ONLY the errors mentioned - do not modify valid components
2. Preserve all existing valid data (question_text, settings, etc.)
3. Output the complete fixed form as valid JSON
4. Ensure all components have: component_id, question_type, question_text
"""
invalid_form = dspy.InputField(desc="The form JSON that failed validation")
validation_errors = dspy.InputField(desc="List of validation errors to fix")
user_query = dspy.InputField(desc="Original user request for context")
fixed_form = dspy.OutputField(desc="Complete fixed form as valid JSON with all errors corrected")
# ============================================================================
# FORM GENERATION MODULE
# ============================================================================
class FormGenerationModule(dspy.Module):
"""Main module for generating forms from natural language."""
# Type correction mapping for common AI mistakes
TYPE_CORRECTIONS = {
'text': 'short_answer',
'textarea': 'long_answer',
'textbox': 'short_answer',
'input': 'short_answer',
'radio': 'multiple_choice',
'radio_button': 'multiple_choice',
'radio_buttons': 'multiple_choice',
'checkbox': 'checkboxes',
'select': 'dropdown',
'select_one': 'dropdown',
'select_multiple': 'multi_select',
'multiselect': 'multi_select',
'file': 'file_upload',
'upload': 'file_upload',
'scale': 'linear_scale',
'slider': 'linear_scale',
'stars': 'rating',
'star_rating': 'rating',
'sign': 'signature',
'rank': 'ranking',
'order': 'ranking',
'wallet': 'wallet_connect',
'web3': 'wallet_connect',
'url': 'link',
'website': 'link',
'tel': 'phone',
'telephone': 'phone',
'datetime': 'date',
'grid': 'matrix',
'table': 'matrix',
'numeric': 'number',
'integer': 'number',
'float': 'number',
'submit': 'button',
'action': 'button',
}
VALID_TYPES = [
'short_answer', 'long_answer', 'multiple_choice', 'checkboxes',
'dropdown', 'multi_select', 'number', 'email', 'phone', 'link',
'file_upload', 'date', 'time', 'linear_scale', 'matrix', 'rating',
'payment', 'signature', 'ranking', 'wallet_connect', 'button'
]
def __init__(self):
self.planner = dspy.Predict(FormPlannerSignature)
self.component_generator = dspy.Predict(ComponentSignatureGenerator)
self.form_fixer = dspy.Predict(FormFixerSignature)
def _correct_question_type(self, question_type: str) -> str:
"""Correct common AI mistakes in question type generation."""
if not question_type:
return 'short_answer'
normalized = question_type.lower().strip().replace(' ', '_').replace('-', '_')
if normalized in self.TYPE_CORRECTIONS:
return self.TYPE_CORRECTIONS[normalized]
if normalized in self.VALID_TYPES:
return normalized
return 'short_answer' # Default fallback
async def aforward(self, user_query: str):
"""
Generate a complete form from a user query.
Args:
user_query: Natural language description of the form
Returns:
dict with complete form structure including components and conditional logic
"""
# Step 1: Generate form plan with component briefs and conditional logic
with dspy.context(lm=dspy.LM("openai/gpt-4o-mini", api_key=os.getenv('OPENAI_API_KEY'), max_tokens=2000)):
plan = self.planner(user_query=user_query)
form_plan = plan.form_plan
if isinstance(form_plan, str):
try:
form_plan = json.loads(form_plan)
except json.JSONDecodeError:
logger.error("Failed to parse form plan JSON")
return {'error': 'Failed to generate form plan'}
logger.info(f"Form plan generated: {form_plan.get('title', 'Untitled')}")
# Step 2: Generate detailed component specifications
components = form_plan.get('components', [])
detailed_components = []
# CRITICAL: Include original user_query so components use exact terminology
form_context = json.dumps({
'title': form_plan.get('title'),
'description': form_plan.get('description'),
'total_components': len(components),
'ORIGINAL_USER_REQUEST': user_query, # Pass the exact user request
'IMPORTANT': 'Use the EXACT terms from ORIGINAL_USER_REQUEST. Do NOT substitute cultural or local terms.'
})
for comp in components:
try:
# Include original user query in component brief for exact terminology
enhanced_brief = f"{comp.get('brief', '')}. IMPORTANT: Use exact terms from user's original request: '{user_query}'. Do NOT substitute any terms."
with dspy.context(lm=dspy.LM("openai/gpt-4o-mini", api_key=os.getenv('OPENAI_API_KEY'), max_tokens=1000)):
component_spec = self.component_generator(
component_brief=enhanced_brief,
component_type=comp.get('type', 'short_answer'),
component_id=comp.get('component_id', f"comp_{len(detailed_components) + 1}"),
form_context=form_context
)
comp_spec = component_spec.component_spec
if isinstance(comp_spec, str):
try:
comp_spec = json.loads(comp_spec)
except json.JSONDecodeError:
logger.error(f"Failed to parse component spec for {comp.get('component_id')}")
continue
# Auto-correct question_type if AI generated an invalid type
if 'question_type' in comp_spec:
original_type = comp_spec['question_type']
corrected_type = self._correct_question_type(original_type)
if corrected_type != original_type:
logger.info(f"Auto-corrected question_type '{original_type}' -> '{corrected_type}'")
comp_spec['question_type'] = corrected_type
# Ensure order is preserved
comp_spec['order'] = comp.get('order', len(detailed_components))
detailed_components.append(comp_spec)
except Exception as e:
logger.error(f"Error generating component {comp.get('component_id')}: {e}")
continue
# Step 3: Build final form structure
final_form = {
'title': form_plan.get('title', 'Untitled Form'),
'description': form_plan.get('description', ''),
'submit_button_text': form_plan.get('submit_button_text', 'Submit'),
'components': detailed_components,
'conditional_logic': form_plan.get('conditional_logic', []),
'settings': form_plan.get('settings', {})
}
return final_form
async def fix_form(self, invalid_form: dict, validation_errors: List[str], user_query: str) -> dict:
"""
Fix validation errors in a form using the FormFixerSignature.
Args:
invalid_form: The form dict that failed validation
validation_errors: List of validation error messages
user_query: Original user request for context
Returns:
Fixed form dict, or the original if fixing fails
"""
try:
logger.info(f"Attempting to fix form with {len(validation_errors)} validation errors")
with dspy.context(lm=dspy.LM("openai/gpt-4o-mini", api_key=os.getenv('OPENAI_API_KEY'), max_tokens=2500)):
fix_result = self.form_fixer(
invalid_form=json.dumps(invalid_form),
validation_errors=json.dumps(validation_errors),
user_query=user_query
)
fixed_form = fix_result.fixed_form
if isinstance(fixed_form, str):
try:
fixed_form = json.loads(fixed_form)
except json.JSONDecodeError:
logger.error("Failed to parse fixed form JSON")
return invalid_form
# Apply type corrections to all components in the fixed form
if 'components' in fixed_form:
for comp in fixed_form['components']:
if 'question_type' in comp:
original_type = comp['question_type']
corrected_type = self._correct_question_type(original_type)
if corrected_type != original_type:
logger.info(f"Post-fix correction: '{original_type}' -> '{corrected_type}'")
comp['question_type'] = corrected_type
logger.info(f"Form fixed successfully")
return fixed_form
except Exception as e:
logger.error(f"Failed to fix form: {e}", exc_info=True)
return invalid_form
# ============================================================================
# UTILITY FUNCTIONS
# ============================================================================
def validate_form_structure(form_data: dict) -> tuple[bool, str]:
"""
Validate form structure.
Args:
form_data: Form dictionary to validate
Returns:
(is_valid, error_message)
"""
if not isinstance(form_data, dict):
return False, "Form data must be a dictionary"
if 'title' not in form_data:
return False, "Form must have a title"
if 'components' not in form_data or not isinstance(form_data['components'], list):
return False, "Form must have a components list"
if len(form_data['components']) == 0:
return False, "Form must have at least one component"
# Validate each component has required fields
for i, comp in enumerate(form_data['components']):
if 'component_id' not in comp:
return False, f"Component {i} missing component_id"
if 'question_type' not in comp:
return False, f"Component {comp.get('component_id')} missing question_type"
if 'question_text' not in comp:
return False, f"Component {comp.get('component_id')} missing question_text"
return True, ""
def extract_component_ids(form_data: dict) -> list:
"""Extract all component IDs from form structure."""
if not isinstance(form_data, dict):
return []
components = form_data.get('components', [])
return [comp.get('component_id') for comp in components if 'component_id' in comp]
# ============================================================================
# RESPONSE ANALYSIS SIGNATURES
# ============================================================================
class ResponseAnalysisRouterSignature(dspy.Signature):
"""Route user queries for form response analysis.
ROUTING RULES:
- **summary**: User wants an overall summary or overview of responses.
Keywords: "summarize", "overview", "tell me about", "how many", "total"
Examples: "Give me a summary of all responses", "How many people responded?"
- **filter**: User wants to filter or search specific responses.
Keywords: "show", "find", "where", "filter", "which", "list"
Examples: "Show me responses where rating > 3", "Find responses from last week"
- **aggregate**: User wants statistical aggregations or calculations.
Keywords: "average", "count", "sum", "max", "min", "distribution", "percentage"
Examples: "What's the average rating?", "How are responses distributed?"
- **sentiment**: User wants sentiment analysis on text responses.
Keywords: "sentiment", "positive", "negative", "feedback", "feel", "opinion"
Examples: "What's the overall sentiment?", "Are responses positive or negative?"
- **export**: User wants to export or download data.
Keywords: "export", "download", "csv", "file", "save"
Examples: "Export responses as CSV", "Download filtered results"
- **general**: General questions about the data or clarification needed.
Examples: "What columns are available?", "Help me understand the data"
"""
user_query = dspy.InputField(desc="The user's query about form responses")
schema_description = dspy.InputField(desc="Description of available data columns")
conversation_history = dspy.InputField(desc="Previous conversation context")
query_type = dspy.OutputField(desc="One of: 'summary', 'filter', 'aggregate', 'sentiment', 'export', 'general'")
reasoning = dspy.OutputField(desc="Brief explanation of why this route was chosen")
class SQLGeneratorSignature(dspy.Signature):
"""Generate safe SQL queries for DuckDB based on user requests.
RULES:
1. Only generate SELECT queries - no INSERT, UPDATE, DELETE, DROP, etc.
2. Use the exact column names from the schema description
3. Use DuckDB SQL syntax (similar to PostgreSQL)
4. For text matching, use ILIKE for case-insensitive search
5. Always include LIMIT clause (max 100 rows) unless counting/aggregating
6. Wrap column names in double quotes if they contain special characters
IMPORTANT - QUESTION SHORTCUTS (q1, q2, etc.):
- The schema includes a SHORTCUTS section mapping q1, q2, etc. to full column names
- When user says "q1", "q2", "question 1", etc., use the SHORTCUTS to find the actual column name
- Example: if schema shows 'q1 β "q1_full_name"', use "q1_full_name" in your SQL
IMPORTANT - NUMERIC COLUMNS (rating, number, linear_scale):
- Values are stored as strings, empty string means NULL
- ALWAYS filter out empty strings BEFORE casting: WHERE "col" != '' AND "col" IS NOT NULL
- For AVG/SUM: SELECT AVG(CAST("col" AS FLOAT)) FROM responses WHERE "col" != '' AND "col" IS NOT NULL
- For comparisons: WHERE "col" != '' AND CAST("col" AS INTEGER) > 3
IMPORTANT - CHECKBOX/MULTI-SELECT COLUMNS:
- Values are comma-separated strings like "Option A, Option B"
- For INDIVIDUAL option counts: use LIKE for each option
- For COMBINATION analysis (Sankey/flow): GROUP BY the entire column to see which combinations are selected together
- Schema will list available options
IMPORTANT - SINGLE CHOICE (multiple_choice, dropdown):
- One value per response, use GROUP BY to count
EXAMPLES:
- "Average of q2" (if q2 β "q2_rating") β SELECT AVG(CAST("q2_rating" AS FLOAT)) as avg_rating FROM responses WHERE "q2_rating" != '' AND "q2_rating" IS NOT NULL
- "Count ratings" β SELECT "q2_rating" as rating, COUNT(*) as count FROM responses WHERE "q2_rating" != '' GROUP BY "q2_rating" ORDER BY rating
- "Ratings above 3" β SELECT * FROM responses WHERE "q2_rating" != '' AND CAST("q2_rating" AS INTEGER) > 3 LIMIT 100
- "Count by satisfaction level" (single choice) β SELECT "q1_satisfaction" as satisfaction, COUNT(*) as count FROM responses WHERE "q1_satisfaction" != '' GROUP BY "q1_satisfaction"
- "Count individual checkbox options" (for checkbox with options A, B, C) β
SELECT 'Option A' as option, COUNT(*) as count FROM responses WHERE "q3_features" ILIKE '%Option A%'
UNION ALL SELECT 'Option B', COUNT(*) FROM responses WHERE "q3_features" ILIKE '%Option B%'
UNION ALL SELECT 'Option C', COUNT(*) FROM responses WHERE "q3_features" ILIKE '%Option C%'
- "Analyze checkbox combinations" or "Show checkbox Sankey" β
SELECT "q3_features" as combination, COUNT(*) as count FROM responses WHERE "q3_features" != '' GROUP BY "q3_features" ORDER BY count DESC
- "Count by status" β SELECT status, COUNT(*) as count FROM responses GROUP BY status
- "Show all responses" β SELECT * FROM responses LIMIT 100
"""
user_query = dspy.InputField(desc="Natural language query from user")
schema_description = dspy.InputField(desc="Table schema with column names, descriptions, QUESTION SHORTCUTS mapping q1/q2 to full column names, and available options for choice fields")
sql_query = dspy.OutputField(desc="Safe SELECT SQL query for DuckDB, using full column names from SHORTCUTS when user references q1/q2")
class ResponseSummarizerSignature(dspy.Signature):
"""Summarize form response data in natural language.
Create a clear, concise summary that highlights:
1. Key statistics (total responses, completion rate)
2. Notable patterns or trends
3. Most common answers for each question
4. Any interesting insights
Keep the summary conversational and easy to understand.
Use bullet points for clarity when listing multiple items.
"""
query_results = dspy.InputField(desc="JSON data from query execution")
user_query = dspy.InputField(desc="What the user asked for")
schema_description = dspy.InputField(desc="Column descriptions for context")
summary = dspy.OutputField(desc="Natural language summary of the data (max 500 words)")
class ResponseChatModule(dspy.Module):
"""Module for handling response analysis chat interactions."""
def __init__(self):
self.router = dspy.Predict(ResponseAnalysisRouterSignature)
self.sql_generator = dspy.Predict(SQLGeneratorSignature)
self.summarizer = dspy.Predict(ResponseSummarizerSignature)
async def aforward(
self,
user_query: str,
schema_description: str,
conversation_history: str,
response_summary: dict = None
) -> dict:
"""
Process a user query about form responses.
Args:
user_query: User's natural language query
schema_description: Description of available columns
conversation_history: Previous conversation context
response_summary: Pre-computed summary statistics
Returns:
dict with 'query_type', 'sql_query' (if applicable), 'response'
"""
# Truncate inputs
user_query = user_query[:1000] if len(user_query) > 1000 else user_query
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=1000, temperature=0.3)):
# Route the query
route_result = self.router(
user_query=user_query,
schema_description=schema_description,
conversation_history=conversation_history
)
query_type = route_result.query_type.lower().strip()
result = {
'query_type': query_type,
'reasoning': route_result.reasoning,
'sql_query': None,
'response': None
}
if query_type in ['filter', 'aggregate', 'export']:
# Generate SQL query
sql_result = self.sql_generator(
user_query=user_query,
schema_description=schema_description
)
result['sql_query'] = sql_result.sql_query
elif query_type == 'summary':
# Use pre-computed summary if available
if response_summary:
summary_context = json.dumps(response_summary, default=str)
summarize_result = self.summarizer(
query_results=summary_context,
user_query=user_query,
schema_description=schema_description
)
result['response'] = summarize_result.summary
elif query_type == 'general':
# Answer general questions about the data
result['response'] = f"I can help you analyze your form responses. The available data includes: {schema_description}\n\nYou can ask me to:\n- Summarize responses\n- Filter by specific criteria\n- Calculate statistics\n- Analyze sentiment of text answers\n- Export filtered data"
return result
def summarize_results(
self,
query_results: list,
user_query: str,
schema_description: str
) -> str:
"""Summarize query results in natural language."""
with dspy.context(lm=dspy.LM('openai/gpt-4o-mini', api_key=os.getenv('OPENAI_API_KEY'), max_tokens=800)):
result = self.summarizer(
query_results=json.dumps(query_results[:50], default=str), # Limit to 50 rows
user_query=user_query,
schema_description=schema_description
)
return result.summary
# Singleton instance for response chat
response_chat_module = ResponseChatModule()
|