Update app.py

app.py

@@ -154,6 +154,7 @@ def parse_llm_response(content: str) -> str:
 @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))
+    print("chunking done")
     if not chunks:
         return JSONResponse({"error": "No documents could be processed."}, status_code=400)
 
@@ -163,23 +164,23 @@ async def run_hackrx(req: RunRequest):
     )
 
     # dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 8})
-    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 10 ,"lambda_mult": 0.5} )
+    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 12 ,"lambda_mult": 0.5} )
     
 
     # Create retrievers using the pre-loaded models from our ml_models dictionary
     keyword_retriever = BM25Retriever.from_documents(chunks)
-    keyword_retriever.k = 6
+    keyword_retriever.k = 8
     # dense_retriever = Chroma.from_documents(documents=chunks, embedding=ml_models["embedder"]).as_retriever()
-    ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.4, 0.6],search_kwargs={"k": 10})
+    ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.4, 0.6],search_kwargs={"k": 11})
     ### to make it faster we are now using our built reranker thats why commenting the code below
-    compression_retriever = ContextualCompressionRetriever(
-        base_retriever=ensemble_retriever, base_compressor=ml_models["reranker_compressor"]
-    )
+    # compression_retriever = ContextualCompressionRetriever(
+    #     base_retriever=ensemble_retriever, base_compressor=ml_models["reranker_compressor"]
+    # )
 
     
     # Define the RAG chain using pre-loaded components
     hybrid_rag_chain = (
-        {"context": itemgetter("full_query") | compression_retriever, "full_query": itemgetter("full_query")}
+        {"context": itemgetter("full_query") | ensemble_retriever, "full_query": itemgetter("full_query")}
         | ml_models["prompt_template"]
         | ml_models["llm"]
     )