Update app.py

app.py

@@ -5,6 +5,8 @@ print("--- Python script starting ---")
 import streamlit as st
 import os
 
+import langchain
+langchain.debug = True 
 os.environ['TOKENIZERS_PARALLELISM'] = 'false'
 os.environ['HF_HOME'] = '/app/huggingface_cache' # For transformers and datasets
 os.environ['TRANSFORMERS_CACHE'] = '/app/huggingface_cache/transformers'
@@ -109,6 +111,47 @@ try:
         input_variables=["context", "question"],
         partial_variables={"format_instructions": format_instructions}
     )
+
+    # --- NEW: Break down the chain for debugging ---
+    def retrieve_and_rerank(input_dict):
+        print(f"--- RAG DEBUG: Retrieving for question: {input_dict['question']} ---")
+        docs = retriever.invoke(input_dict['question'])
+        print(f"--- RAG DEBUG: Retrieved {len(docs)} docs after reranking ---")
+        for i, doc in enumerate(docs):
+            print(f"    Doc {i} (source: {doc.metadata.get('source', 'N/A')}, page: {doc.metadata.get('page', 'N/A')}): {doc.page_content[:100]}...")
+        return {"context": docs, "question": input_dict['question']}
+    
+    def format_prompt(input_dict):
+        print(f"--- RAG DEBUG: Formatting prompt with context ---")
+        # Manually construct the context string to see it clearly
+        context_str = "\n\n---\n\n".join([doc.page_content for doc in input_dict['context']])
+        print(f"--- RAG DEBUG: Context fed to LLM: {context_str[:500]}... ---") # Print first 500 chars of context
+        return prompt.invoke({"context": context_str, "question": input_dict['question']})
+    
+    def call_llm(formatted_prompt):
+        print(f"--- RAG DEBUG: Calling LLM ---")
+        llm_output = llm.invoke(formatted_prompt)
+        print(f"--- RAG DEBUG: Raw LLM Output: {llm_output} ---") # See exactly what Groq returns
+        return llm_output
+    
+    def parse_output(llm_output_str):
+        print(f"--- RAG DEBUG: Attempting to parse LLM output with Pydantic ---")
+        try:
+            parsed = pydantic_parser.invoke(llm_output_str)
+            print(f"--- RAG DEBUG: Pydantic parsing successful ---")
+            return parsed
+        except Exception as e_parse:
+            print(f"!!!!!!!!!! PYDANTIC PARSING ERROR !!!!!!!!!!")
+            print(f"Raw LLM Output that failed to parse: {llm_output_str}")
+            print(traceback.format_exc())
+            # Fallback: return a dictionary indicating failure, or just the raw string
+            return StructuredAnswer(summary="LLM output parsing failed. See logs.", key_points=[], confidence_score=0.0)
+
+
+
+
+
+    
     rag_chain = (
         {"context": retriever, "question": RunnablePassthrough()}
         | prompt