Update app.py

app.py

@@ -207,10 +207,16 @@ async def run_hackrx(req: RunRequest):
         docs = await ensemble_retriever.ainvoke(question)
         doc_texts = [doc.page_content for doc in docs]
         doc_embeddings = ml_models["embedder"].embed_documents(doc_texts)
-        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]
+        # sims = cosine_similarity([question_embeddings[q_idx]], doc_embeddings)[0]
+        query_embedding = question_embeddings[q_idx]
+        selected_indices = mmr_select(
+        query_embedding=query_embedding,
+        candidates=doc_embeddings,
+        k=6,
+        lambda_mult=0.6,
+    )
+        # top_indices = np.argsort(sims)[-top_k:][::-1]
+        top_chunks = [doc_texts[j] for j in selected_indices]
         return "\n\n".join(top_chunks)
     # Retrieve and rerank all in parallel
     retrieved_chunks_all = await asyncio.gather(