import os import json import asyncio from fastapi import HTTPException from langchain_groq import ChatGroq from langchain_google_genai import ChatGoogleGenerativeAI # Kept for future fallback! from langchain_core.prompts import PromptTemplate async def extract_entities(text: str) -> list[dict]: # 1. Load keys (Keeping both in the codebase as requested) groq_key = os.getenv("GROQ_API_KEY") google_key = os.getenv("GOOGLE_API_KEY") # Kept for fallback if not groq_key: raise HTTPException(status_code=500, detail="Groq API Key missing in .env file.") try: print("\n--- EXTRACTING ENTITIES WITH GROQ ---") # 2. Connect to Groq llm = ChatGroq( model="openai/gpt-oss-120b", api_key=groq_key, temperature=0.0 # Strict zero temperature for formatting consistency ) # 3. Prompt tweaked to ensure strict JSON output for Llama 3 template = """ Analyze the following text and extract the key fantasy entities (characters, locations, events, or organizations). CRITICAL INSTRUCTIONS: 1. Return the result ONLY as a valid JSON array of objects. 2. Absolutely NO conversational text, NO markdown formatting, and NO ```json blocks. 3. Each object must have exactly two keys: "name" (the entity name) and "type" (must be "character", "location", "event", or "organization"). Text: {text} """ prompt = PromptTemplate.from_template(template) chain = prompt | llm # 4. Get the response response = await chain.ainvoke({"text": text}) # 5. Clean up in case Llama 3 disobeys the "no markdown" rule raw_text = response.content.strip() if raw_text.startswith("```json"): raw_text = raw_text[7:-3].strip() elif raw_text.startswith("```"): raw_text = raw_text[3:-3].strip() entities = json.loads(raw_text) print(f"Success! Found {len(entities)} entities.") print("-------------------------------------\n") return entities except Exception as e: print(f"\n--- EXTRACTION ERROR ---") print(str(e)) print("------------------------\n") # Return an empty list if it fails so the frontend doesn't crash return []