Update app.py

app.py

@@ -14,8 +14,8 @@ from fastapi.responses import JSONResponse
 # Make sure you have these files in a 'utils' folder
 from utils.DocsLoader import load_and_chunk
 from utils.Schemas import RunRequest, RunResponse
-from concurrent.futures import ThreadPoolExecutor
-from langchain.vectorstores import FAISS
+# from concurrent.futures import ThreadPoolExecutor
+from langchain_community.vectorstores import FAISS
 from langchain.schema import Document
 from langchain_google_genai import ChatGoogleGenerativeAI
 from langchain_huggingface import HuggingFaceEmbeddings # Correct new import
@@ -143,39 +143,31 @@ def parse_llm_response(content: str) -> str:
 # --- 5. Main API Endpoint ---
 @app.post("/api/v1/hackrx/run", response_model=RunResponse, dependencies=[Depends(verify_api_key)])
 async def run_hackrx(req: RunRequest):
-    chunks = load_and_chunk(str(req.documents))
-    if not chunks:
-        return JSONResponse({"error": "No documents could be processed."}, status_code=400)
+    # chunks = load_and_chunk(str(req.documents))
+    # if not chunks:
+    #     return JSONResponse({"error": "No documents could be processed."}, status_code=400)
 
         
     ####code for parallel####################################################################################
+    all_chunks = []
+    for doc_url in req.documents:
+        chunks = load_and_chunk(doc_url)
+        all_chunks.extend(chunks)
+
+    if not all_chunks:
+        return JSONResponse({"error": "No documents could be processed."}, status_code=400)
+
+    vectorstore = await FAISS.afrom_documents(
+        documents=all_chunks, 
+        embedding=ml_models["embedder"]
+    )
 
-    
-    docs = [Document(page_content=d.page_content) for d in chunks]
-
-    # Parallel embedding
-    def embed_batch(batch):
-        texts = [d.page_content for d in batch]
-        embs = ml_models['embedder'].embed_documents(texts)
-        for doc, emb in zip(batch, embs):
-            doc.embedding = emb
-        return batch
-
-    # Batch and run
-    batch_size = 64
-    batches = [docs[i:i+batch_size] for i in range(0, len(docs), batch_size)]
-    embedded_docs = []
-    with ThreadPoolExecutor() as executor:
-        for batch in executor.map(embed_batch, batches):
-            embedded_docs.extend(batch)
-
-    vectorstore = FAISS.from_documents(embedded_docs, embedding=None)
     dense_retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
     
 
     # Create retrievers using the pre-loaded models from our ml_models dictionary
-    keyword_retriever = BM25Retriever.from_documents(chunks)
-    keyword_retriever.k = 4
+    keyword_retriever = BM25Retriever.from_documents(all_chunks)
+    keyword_retriever.k = 3
     # dense_retriever = Chroma.from_documents(documents=chunks, embedding=ml_models["embedder"]).as_retriever()
     ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.4, 0.65])