Update app.py

app.py

@@ -188,21 +188,34 @@ async def run_hackrx(req: RunRequest):
     question_embeddings = ml_models["embedder"].embed_documents(req.questions)
 
     # For each question, retrieve and rerank with cosine
-    retrieved_chunks_all = []
-    for i, question in enumerate(req.questions):
+    # retrieved_chunks_all = []
+    # for i, question in enumerate(req.questions):
+    #     docs = ensemble_retriever.get_relevant_documents(question)
+    #     doc_texts = [doc.page_content for doc in docs]
+
+    #     doc_embeddings = ml_models["embedder"].embed_documents(doc_texts)
+    #     sims = cosine_similarity([question_embeddings[i]], doc_embeddings)[0]
+
+    #     top_k = 5
+    #     top_indices = np.argsort(sims)[-top_k:][::-1]
+    #     top_chunks = [doc_texts[j] for j in top_indices]
+
+    #     # Join for context
+    #     joined_context = "\n\n".join(top_chunks)
+    #     retrieved_chunks_all.append(joined_context)
+    async def async_retrieve_and_rerank(question: str, q_idx: int):
         docs = ensemble_retriever.get_relevant_documents(question)
         doc_texts = [doc.page_content for doc in docs]
-
         doc_embeddings = ml_models["embedder"].embed_documents(doc_texts)
-        sims = cosine_similarity([question_embeddings[i]], doc_embeddings)[0]
-
-        top_k = 5
+        sims = cosine_similarity([question_embeddings[q_idx]], doc_embeddings)[0]
+        top_k = 6
         top_indices = np.argsort(sims)[-top_k:][::-1]
         top_chunks = [doc_texts[j] for j in top_indices]
-
-        # Join for context
-        joined_context = "\n\n".join(top_chunks)
-        retrieved_chunks_all.append(joined_context)
+        return "\n\n".join(top_chunks)
+    # Retrieve and rerank all in parallel
+    retrieved_chunks_all = await asyncio.gather(
+        *[async_retrieve_and_rerank(q, i) for i, q in enumerate(req.questions)]
+    ) 
 
     ####################################################################################################################