Update app.py

app.py

@@ -10,7 +10,9 @@ from langchain.prompts import PromptTemplate
 from transformers import AutoModelForSeq2SeqLM, pipeline, AutoTokenizer
 import torch
 
-# Check if about_me.txt exists, create a sample if not
+# -------------------------------
+# 1. Ensure about_me.txt exists
+# -------------------------------
 if not os.path.exists("about_me.txt"):
     with open("about_me.txt", "w") as f:
         f.write("""
@@ -18,37 +20,44 @@ if not os.path.exists("about_me.txt"):
         This is a sample portfolio text. Please replace this with your actual portfolio content.
         """)
 
-# Load data
+# -------------------------------
+# 2. Load data
+# -------------------------------
 try:
     loader = TextLoader("about_me.txt")
     docs = loader.load()
 except Exception as e:
     print(f"Error loading document: {e}")
-    # Create fallback document
     from langchain.schema import Document
     docs = [Document(page_content="Hello! I am a portfolio chatbot ready to help you.")]
 
-# Split documents
+# -------------------------------
+# 3. Split documents into chunks
+# -------------------------------
 text_splitter = RecursiveCharacterTextSplitter(
-    chunk_size=200,  # Reduced chunk size
-    chunk_overlap=30
+    chunk_size=500,     # Larger chunk size for better context
+    chunk_overlap=50
 )
 split_docs = text_splitter.split_documents(docs)
 
-# Initialize embeddings and vector store
+# -------------------------------
+# 4. Create embeddings & FAISS DB
+# -------------------------------
 print("Loading embeddings...")
 embedding_model = HuggingFaceEmbeddings(
     model_name="sentence-transformers/all-MiniLM-L6-v2",
-    model_kwargs={'device': 'cpu'}  # Force CPU usage
+    model_kwargs={'device': 'cpu'}
 )
 
 print("Creating vector database...")
 db = FAISS.from_documents(split_docs, embedding_model)
 
-# Load smaller model with better error handling
+# -------------------------------
+# 5. Load language model
+# -------------------------------
 print("Loading language model...")
-model_id = "google/flan-t5-small"  # Changed to smaller model
-device = "cpu"  # Force CPU usage for stability
+model_id = "google/flan-t5-base"  # More capable than small
+device = "cpu"
 
 try:
     tokenizer = AutoTokenizer.from_pretrained(model_id)
@@ -57,77 +66,92 @@ try:
         torch_dtype=torch.float32,
         device_map="auto" if torch.cuda.is_available() else None
     )
-    
+
     pipe = pipeline(
         "text2text-generation",
         model=model,
         tokenizer=tokenizer,
-        max_new_tokens=128,  # Only use max_new_tokens to avoid warning
+        max_new_tokens=128,
         truncation=True,
         device=0 if torch.cuda.is_available() else -1
     )
-    
+
     llm = HuggingFacePipeline(pipeline=pipe)
     print("Model loaded successfully!")
-    
+
 except Exception as e:
     print(f"Error loading model: {e}")
-    # Fallback to a simpler setup
-    from langchain_community.llms import HuggingFacePipeline
     pipe = pipeline(
         "text-generation",
         model="microsoft/DialoGPT-medium",
         max_length=200,
-        device=-1  # CPU only
+        device=-1
     )
     llm = HuggingFacePipeline(pipeline=pipe)
 
-# Create RetrievalQA chain
+# -------------------------------
+# 6. Custom Prompt
+# -------------------------------
 custom_prompt = PromptTemplate(
-    template="Based on the following context, answer the question concisely:\n\nContext: {context}\n\nQuestion: {question}\n\nAnswer:",
+    template=(
+        "Answer the question using only the provided context. "
+        "If the answer is not in the context, say you don't know.\n\n"
+        "Question: {question}\n\n"
+        "Context: {context}\n\n"
+        "Answer:"
+    ),
     input_variables=["context", "question"]
 )
 
+# -------------------------------
+# 7. Create RetrievalQA chain
+# -------------------------------
 qa_chain = RetrievalQA.from_chain_type(
     llm=llm,
     chain_type="stuff",
-    retriever=db.as_retriever(search_kwargs={"k": 1}),  # Reduced to 1 document
+    retriever=db.as_retriever(search_kwargs={"k": 3}),  # Fetch more context
     chain_type_kwargs={"prompt": custom_prompt},
     return_source_documents=False
 )
 
+# -------------------------------
+# 8. Ask function with debug logs
+# -------------------------------
 def ask_bot_alternative(question):
-    """Enhanced chatbot function with better error handling"""
     try:
-        if not question or question.strip() == "":
+        if not question.strip():
             return "Please ask me a question about the portfolio!"
-        
-        # Limit input length
+
         question = question[:500]
-        
-        print(f"Processing question: {question}")
+        print(f"\nProcessing question: {question}")
+
+        # Retrieve and log context
+        retriever = db.as_retriever(search_kwargs={"k": 3})
+        context_docs = retriever.get_relevant_documents(question)
+        print("\n--- Retrieved Context ---")
+        for i, d in enumerate(context_docs, 1):
+            print(f"[Doc {i}] {d.page_content[:200]}...\n")
+
+        # Get answer from chain
         response = qa_chain.invoke({"query": question})
-        
-        # Extract answer with multiple fallbacks
         if isinstance(response, dict):
             answer = response.get("result") or response.get("answer") or str(response)
         else:
             answer = str(response)
-        
-        # Clean and limit output
+
         answer = answer.strip()
         if len(answer) > 1000:
             answer = answer[:1000] + "..."
-            
-        return answer if answer else "I couldn't generate a response. Please try rephrasing your question."
-        
+
+        return answer or "I couldn't find an answer in the portfolio content."
+
     except Exception as e:
         print(f"Error in ask_bot_alternative: {e}")
-        return f"Sorry, I encountered an error: {str(e)[:200]}. Please try again with a different question."
-
-# Create Gradio interface with better configuration
-print("Starting Gradio interface...")
+        return f"Sorry, I encountered an error: {str(e)[:200]}"
 
+# -------------------------------
+# 9. Gradio Interface
+# -------------------------------
 iface = gr.Interface(
     fn=ask_bot_alternative,
     inputs=gr.Textbox(
@@ -156,4 +180,4 @@ if __name__ == "__main__":
         server_name="0.0.0.0",
         server_port=7860,
         show_error=True
-    )
+    )