Update app.py

app.py

@@ -10,6 +10,7 @@ from dotenv import load_dotenv
 from operator import itemgetter
 from fastapi import FastAPI, Depends, HTTPException, Header
 from fastapi.responses import JSONResponse
+from langchain_core.documents import Document
 from utils.DocsLoader import load_and_chunk
 from utils.Schemas import RunRequest, RunResponse
 from langchain_community.vectorstores import FAISS
@@ -19,6 +20,7 @@ from langchain_community.retrievers import BM25Retriever
 from langchain.retrievers import EnsembleRetriever
 from sklearn.metrics.pairwise import cosine_similarity
 from langchain.prompts import PromptTemplate
+from fastapi.middleware.cors import CORSMiddleware  # Optional, for external frontend
 
 # Load environment variables
 load_dotenv()
@@ -39,7 +41,6 @@ async def lifespan(app: FastAPI):
         if not GOOGLE_API_KEY:
             raise RuntimeError("CRITICAL: Missing GOOGLE_API_KEY in environment secrets!")
     
-        # Optimized embedding model with batching
         ml_models["embedder"] = HuggingFaceEmbeddings(
             model_name="BAAI/bge-base-en-v1.5",
             encode_kwargs={
@@ -48,7 +49,6 @@ async def lifespan(app: FastAPI):
             }
         )
         
-        # Faster LLM with constrained output
         ml_models["llm"] = ChatGoogleGenerativeAI(
             model="gemini-1.5-pro",
             api_key=GOOGLE_API_KEY,
@@ -56,7 +56,6 @@ async def lifespan(app: FastAPI):
             max_output_tokens=300
         )
         
-        # Improved prompt template
         ml_models["prompt_template"] = PromptTemplate.from_template("""
 **Role**: Insurance Policy Expert
 **Context**: 
@@ -85,6 +84,7 @@ async def lifespan(app: FastAPI):
 # --- 2. FastAPI App Instance ---
 app = FastAPI(title="HackRX RAG Server", lifespan=lifespan)
 
+
 # --- 3. API Key Verification ---
 TEAM_API_KEY = os.getenv("TEAM_API_KEY")
 
@@ -98,7 +98,6 @@ def verify_api_key(authorization: str = Header(...)):
 # --- 4. Parsing Helper ---
 def parse_llm_response(content: str) -> str:
     try:
-        # Clean JSON response
         content_cleaned = re.sub(r"^```json|```$", "", content.strip(), flags=re.IGNORECASE).strip()
         data = json.loads(content_cleaned)
         
@@ -118,7 +117,6 @@ def parse_llm_response(content: str) -> str:
 
 # --- 5. Retrieval Optimization ---
 async def get_relevant_docs(question: str, vectorstore: FAISS, keyword_retriever: BM25Retriever):
-    # Parallel retrieval
     dense_docs, sparse_docs = await asyncio.gather(
         vectorstore.asimilarity_search(question, k=6),
         asyncio.get_event_loop().run_in_executor(
@@ -128,11 +126,9 @@ async def get_relevant_docs(question: str, vectorstore: FAISS, keyword_retriever
         )
     )
     
-    # Combine and deduplicate
     all_docs = dense_docs + sparse_docs
     unique_docs = {doc.page_content: doc for doc in all_docs}.values()
     
-    # Fast reranking
     query_embedding = ml_models["embedder"].embed_query(question)
     doc_texts = [doc.page_content for doc in unique_docs]
     doc_embeddings = ml_models["embedder"].embed_documents(doc_texts)
@@ -147,14 +143,16 @@ async def get_relevant_docs(question: str, vectorstore: FAISS, keyword_retriever
 async def run_hackrx(req: RunRequest):
     start_time = time.time()
     
-    # Cache document processing
     async with cache_lock:
-        if req.documents in document_cache:
+        doc_key = hash(str(req.documents))  # Hash key for safety
+        if doc_key in document_cache:
             print("♻️ Using cached document")
-            chunks = document_cache[req.documents]
+            chunks = document_cache[doc_key]
         else:
-            chunks = load_and_chunk(str(req.documents))
-            document_cache[req.documents] = chunks
+            raw_chunks = load_and_chunk(str(req.documents))
+            # 🔧 FIXED: Convert string chunks to Document objects
+            chunks = [Document(page_content=chunk) for chunk in raw_chunks]
+            document_cache[doc_key] = chunks
     
     if not chunks:
         return JSONResponse({"error": "No documents processed"}, status_code=400)
@@ -169,7 +167,6 @@ async def run_hackrx(req: RunRequest):
         relevant_docs = await get_relevant_docs(q, vectorstore, keyword_retriever)
         context = "\n".join([d.page_content for d in relevant_docs])
         
-        # Generate response
         prompt = ml_models["prompt_template"].format_prompt(
             full_query=q, 
             context=context
@@ -179,7 +176,6 @@ async def run_hackrx(req: RunRequest):
     
     answers = await asyncio.gather(*(process_question(q) for q in req.questions))
     
-    # Performance logging
     proc_time = time.time() - start_time
     print(f"⏱️ Processed {len(req.questions)} questions in {proc_time:.2f}s")
     
@@ -187,4 +183,4 @@ async def run_hackrx(req: RunRequest):
 
 @app.get("/", include_in_schema=False)
 def root():
-    return {"message": "HackRX API operational. Use /api/v1/hackrx/run"}
+    return {"message": "HackRX API operational. Use /api/v1/hackrx/run"}