backend/services/question_handler.py

from sqlalchemy.orm import Session
from services.vector_store import VectorStore
from services.llm_service import generate_response, generate_quiz
from services.cache_manager import CacheManager
from models.document import Document

class QuestionHandler:
    def __init__(self):
        self.vector_store = VectorStore()
        self.cache = CacheManager()

    def get_answer(self, question: str, session_id: str, db: Session, socratic_mode: bool = False):
        cache_key = f"{session_id}::{question}::{socratic_mode}"
        cached_answer = self.cache.get(cache_key)
        if cached_answer:
            return {**cached_answer, "cached": True}

        search_results = self.vector_store.find_similar_chunks(
            question=question,
            session_id=session_id,
            n_results=5
        )

        if not search_results or not search_results.get("documents") or not search_results["documents"][0]:
            return {
                "question": question,
                "answer": "Désolé, je n'ai trouvé aucune information pertinente dans les documents fournis. Assurez-vous qu'un document a bien été chargé pour cette session.",
                "sources": [],
                "cached": False
            }

        documents_content = search_results["documents"][0]
        metadatas = search_results["metadatas"][0]

        context_parts = []
        for i, doc_text in enumerate(documents_content):
            meta = metadatas[i]
            filename = meta.get("filename", "Document inconnu")
            page = meta.get("page", "?")
            context_parts.append(f"--- SOURCE : {filename} (Page {page}) ---\nCONTENU : {doc_text}")

        full_context = "\n\n".join(documents_content)
        
        answer = generate_response(question, full_context, socratic_mode)

        sources = []
        doc_ids = {meta.get("document_id") for meta in metadatas if meta.get("document_id") is not None}
        if doc_ids and db:
            db_docs = db.query(Document).filter(Document.id.in_(doc_ids)).all()
            db_titles = {doc.id: doc.file_name for doc in db_docs}
        else:
            db_titles = {}

        seen_sources = set()
        for meta in metadatas:
            doc_id = meta.get("document_id")
            final_title = db_titles.get(doc_id, meta.get("filename", "Document inconnu"))
            page = meta.get("page")
            key = (final_title, page)
            if key not in seen_sources:
                seen_sources.add(key)
                sources.append({"document": final_title, "page": page, "doc_id": doc_id})

        final_response = {
            "question": question,
            "answer": answer,
            "sources": sources
        }
        self.cache.set(cache_key, final_response)
        return {**final_response, "cached": False}

    def get_quiz(self, session_id: str, db: Session):
        import random
        
        search_results = self.vector_store.find_similar_chunks(
            question="concepts importants et points cles",
            session_id=session_id,
            n_results=30
        )

        if not search_results or not search_results.get("documents") or not search_results["documents"][0]:
            return {"quiz": '{"questions": []}'}

        documents_content = search_results["documents"][0]
        random.shuffle(documents_content)
        sampled_content = documents_content[:5]
        full_context = "\n\n".join(sampled_content)

        quiz_json = generate_quiz(full_context)
        
        final_response = {"quiz": quiz_json}
        return final_response