139 lines
5.8 KiB
Python
139 lines
5.8 KiB
Python
"""
|
|
Prompt builder for document generation.
|
|
This module builds prompts for generating documents from extracted content.
|
|
"""
|
|
|
|
import logging
|
|
from typing import Dict, Any
|
|
from modules.datamodels.datamodelJson import jsonTemplateDocument
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
|
|
async def buildGenerationPrompt(
|
|
outputFormat: str,
|
|
userPrompt: str,
|
|
title: str,
|
|
extracted_content: str = None,
|
|
continuationContext: Dict[str, Any] = None
|
|
) -> str:
|
|
"""
|
|
Build the unified generation prompt using a single JSON template.
|
|
Generic solution that works for any user request.
|
|
|
|
Args:
|
|
outputFormat: Target output format (html, pdf, docx, etc.) - not used in prompt
|
|
userPrompt: User's original prompt for document generation
|
|
title: Title for the document
|
|
extracted_content: Optional extracted content from documents to prepend to prompt
|
|
continuationContext: Optional context from previous generation for continuation
|
|
|
|
Returns:
|
|
Complete generation prompt string
|
|
"""
|
|
# Create a template - let AI generate title if not provided
|
|
titleValue = title if title else "Generated Document"
|
|
jsonTemplate = jsonTemplateDocument.replace("{{DOCUMENT_TITLE}}", titleValue)
|
|
|
|
# Build prompt based on whether this is a continuation or first call
|
|
# Check if we have valid continuation context with actual JSON fragment
|
|
hasContinuation = (
|
|
continuationContext
|
|
and continuationContext.get("section_count", 0) > 0
|
|
and continuationContext.get("last_raw_json", "")
|
|
and continuationContext.get("last_raw_json", "").strip() != "{}"
|
|
)
|
|
|
|
if hasContinuation:
|
|
# CONTINUATION PROMPT - user already received first part, continue from where it stopped
|
|
lastRawJson = continuationContext.get("last_raw_json", "")
|
|
lastItemObject = continuationContext.get("last_item_object", "") # Full object like {"text": "value"}
|
|
lastItemsFromFragment = continuationContext.get("last_items_from_fragment", "")
|
|
totalItemsCount = continuationContext.get("total_items_count", 0)
|
|
|
|
# Show the last few items to indicate where to continue (limit fragment size)
|
|
# Extract just the ending portion of the JSON to show where it cut off
|
|
fragmentSnippet = ""
|
|
if lastRawJson:
|
|
# Show last 1500 chars or the whole thing if shorter - just enough to show the cut point
|
|
fragmentSnippet = lastRawJson[-1500:] if len(lastRawJson) > 1500 else lastRawJson
|
|
# Add ellipsis if truncated
|
|
if len(lastRawJson) > 1500:
|
|
fragmentSnippet = "..." + fragmentSnippet
|
|
|
|
# Build clear continuation guidance
|
|
continuationGuidance = []
|
|
|
|
if totalItemsCount > 0:
|
|
continuationGuidance.append(f"You have already generated {totalItemsCount} items.")
|
|
|
|
# Show the last complete item object (full object format)
|
|
if lastItemObject:
|
|
continuationGuidance.append(f"Last item in previous response: {lastItemObject}. Continue with the NEXT item after this.")
|
|
|
|
continuationText = "\n".join(continuationGuidance) if continuationGuidance else "Continue from where it stopped."
|
|
|
|
# PROMPT FOR CONTINUATION
|
|
|
|
generationPrompt = f"""User request: "{userPrompt}"
|
|
|
|
The user already received part of the response. Continue generating the remaining content.
|
|
|
|
{continuationText}
|
|
|
|
Previous response ended here (JSON was cut off at this point):
|
|
```json
|
|
{fragmentSnippet if fragmentSnippet else "(No fragment available)"}
|
|
```
|
|
|
|
JSON structure template:
|
|
{jsonTemplate}
|
|
|
|
Instructions:
|
|
- Return ONLY valid JSON (strict). No comments of any kind (no //, /* */, or #). No trailing commas. Strings must use double quotes.
|
|
- Arrays must contain ONLY JSON values; do not include comments or ellipses.
|
|
- Use ONLY the element structures shown in the template.
|
|
- Continue from where it stopped — add NEW items only; do not repeat existing items.
|
|
- Generate remaining content to complete the user request. Do NOT just give an instruction or comments. Deliver the complete response.
|
|
- Fill with actual content (no placeholders or instructional text such as "Add more...").
|
|
- IMPORTANT: Ensure "filename" in each document has meaningful name with appropriate extension matching the content.
|
|
- When the request is fully satisfied, add "complete_response": true at root level.
|
|
- Output JSON only; no markdown fences or extra text.
|
|
|
|
IMPORTANT: Before responding, analyse the remaining data to fully satisfy user request.
|
|
|
|
Continue generating:
|
|
"""
|
|
else:
|
|
|
|
# PROMPT FOR FIRST CALL
|
|
|
|
generationPrompt = f"""User request: "{userPrompt}"
|
|
|
|
Generate a VALID JSON response for the user request. The template below shows ONLY the structure pattern - it is NOT existing content.
|
|
|
|
JSON structure template:
|
|
{jsonTemplate}
|
|
|
|
Instructions:
|
|
- Start with {{"metadata": ...}} — return COMPLETE, STRICT JSON.
|
|
- Return ONLY valid JSON (strict). No comments. No trailing commas. Use double quotes.
|
|
- Do NOT reuse example section IDs; create your own.
|
|
- Generate complete content based on the user request. Do NOT just give an instruction or comments. Deliver the complete response.
|
|
- IMPORTANT: Set a meaningful "filename" in each document with appropriate file extension (e.g., "prime_numbers.txt", "report.docx", "data.json"). The filename should reflect the content and task objective.
|
|
- When the request is fully satisfied, add "complete_response": true at root level.
|
|
- Output JSON only; no markdown fences or extra text.
|
|
|
|
Generate your complete response starting from {{"metadata": ...}}:
|
|
"""
|
|
|
|
# If we have extracted content, prepend it to the prompt
|
|
if extracted_content:
|
|
generationPrompt = f"""EXTRACTED CONTENT FROM DOCUMENTS:
|
|
{extracted_content}
|
|
|
|
{generationPrompt}"""
|
|
|
|
return generationPrompt.strip()
|
|
|