Update app.py

app.py

@@ -227,9 +227,6 @@ def translate_text(text, source_to_target='ar_to_en'):
 def embed_query_text(query_text):
     query_embedding = embedding.encode([query_text])
     return query_embedding
-    
-from sklearn.metrics.pairwise import cosine_similarity
-import numpy as np
 
 def query_embeddings(query_embedding, embeddings_data=None, n_results=5):
     embeddings_data = embeddings_data or data.get('embeddings', {})
@@ -446,20 +443,45 @@ async def health_check():
 async def chat_endpoint(chat_query: ChatQuery):
     try:
         query_text = chat_query.query
+        
+        # Step 1: Embed the query
         query_embedding = embed_query_text(query_text)
+        
+        # Step 2: Retrieve top results using embeddings similarity
         initial_results = query_embeddings(query_embedding, embeddings_data, n_results=5)
         document_ids = [doc_id for doc_id, _ in initial_results]
+        
+        # Step 3: Fetch document texts
         document_texts = retrieve_document_texts(document_ids, folder_path)
+        
+        # Step 4: Re-rank documents (optional, if reranking is used)
+        reranked_documents = rerank_documents(query_text, document_ids, document_texts, cross_encoder_model)
+        
+        # Step 5: Extract relevant portions (if enabled)
+        relevant_portions = extract_relevant_portions(
+            document_texts,
+            query=query_text,
+            max_portions=3,
+            portion_size=1,
+            min_query_words=1
+        )
+        
+        # Step 6: Flatten and clean relevant portions
         flattened_relevant_portions = []
         for doc_id, portions in relevant_portions.items():
             flattened_relevant_portions.extend(portions)
         unique_selected_parts = remove_duplicates(flattened_relevant_portions)
         combined_parts = " ".join(unique_selected_parts)
-        context = [query_text] + unique_selected_parts
+        
+        # Step 7: Extract entities and enhance passage
         entities = extract_entities(query_text)
         passage = enhance_passage_with_entities(combined_parts, entities)
+        
+        # Step 8: Create prompt and generate answer
         prompt = create_prompt(query_text, passage)
         answer, generation_time = generate_answer(prompt)
+        
+        # Step 9: Clean the generated answer
         answer_part = answer.split("Answer:")[-1].strip()
         cleaned_answer = remove_answer_prefix(answer_part)
         final_answer = remove_incomplete_sentence(cleaned_answer)
@@ -469,9 +491,11 @@ async def chat_endpoint(chat_query: ChatQuery):
             "conversation_id": chat_query.conversation_id,
             "success": True
         }
+
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
+
 @app.post("/api/resources")
 async def resources_endpoint(profile: MedicalProfile):
     try: