added deploment code

Author: RATED-R-SUNDRAM

.github/workflows/sqlite_db

@@ -5,24 +5,29 @@
 from langchain_pinecone import PineconeVectorStore
 from langchain.prompts import PromptTemplate
 from langchain_openai import ChatOpenAI
-from langchain_core.runnables import RunnablePassthrough
 from pinecone import Pinecone, ServerlessSpec
 
 # Load environment variables
 load_dotenv()
 pinecone_api_key = st.secrets["PINECONE_API_KEY2"]
 openai_api_key = st.secrets["OPENAI_API_KEY"]
 
-# Session State Initialization (Moved to top)
+# Session State Initialization
 if "history" not in st.session_state:
     st.session_state.history = []
 if "last_chunks" not in st.session_state:
     st.session_state.last_chunks = []
 
 # Embeddings and LLM
-embedding_model = HuggingFaceEmbeddings(
-    model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={"device": "cpu"}
-)
+try:
+    embedding_model = HuggingFaceEmbeddings(
+        model_name="sentence-transformers/all-MiniLM-L6-v2",
+        model_kwargs={"device": "cpu"},  # Explicitly set device
+    )
+except Exception as e:
+    st.error(f"Error initializing embeddings: {e}. Please ensure 'sentence-transformers' is installed with `pip install sentence-transformers` and compatible PyTorch version.")
+    embedding_model = None
+
 llm = ChatOpenAI(
     model="gpt-4o-mini",
     temperature=0.2,
@@ -33,7 +38,7 @@
 # Pinecone Initialization
 pc = Pinecone(api_key=pinecone_api_key)
 index_name = "chat-models-v1-all-minilm-l6"
-embedding_dim = len(embedding_model.embed_query("test"))
+embedding_dim = len(embedding_model.embed_query("test")) if embedding_model else 384  # Default dim if embedding fails
 
 if index_name not in pc.list_indexes().names():
     pc.create_index(
@@ -45,72 +50,23 @@
 
 vector_store = PineconeVectorStore(
     index_name=index_name, embedding=embedding_model, pinecone_api_key=pinecone_api_key
-)
-retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 6})
-
-# Medical Abbreviations
-MEDICAL_ABBRS = {
-    "IUFD": "Intrauterine fetal death",
-    "PROM": "Prelabour Rupture of Membranes"
-    # Add more as needed
-}
-
-# Intent Examples
-INTENT_EXAMPLES = [
-    {"question": "What are symptoms?", "intent": "new_question"},
-    {"question": "Explain more", "intent": "follow_up"},
-    {"question": "What about diabetes?", "intent": "follow_up"},
-    {"question": "Thanks!", "intent": "gratitude"},
-    {"question": "How are you?", "intent": "chitchat"}
-]
-
-intent_prompt = PromptTemplate.from_template(
-    """
-You are an intent classifier for a medical chatbot. Classify the user's message into:
-- new_question: A fresh medical query.
-- follow_up: Seeks elaboration or new info based on prior context.
-- gratitude: Thanks or positive feedback.
-- chitchat: Casual talk.
-
-Output one label: new_question, follow_up, gratitude, or chitchat.
-
-Examples:
-{% for ex in examples %}- {{ ex.question }} → {{ ex.intent }}
-{% endfor %}
-
-User: {input}
-Intent:
-"""
-)
-
-def build_intent_prompt(user_input: str, examples: list[dict]) -> str:
-    formatted_examples = "\n".join([f"- {ex['question']} → {ex['intent']}" for ex in examples])
-    history = "\n".join([f"User: {q} | Bot: {a}" for q, a, _ in st.session_state.history[-2:]])
-    return f"""
-You are an intent classifier. Classify based on recent history:
-- new_question: Fresh query.
-- follow_up: Elaboration or new info from prior context.
-- gratitude: Thanks.
-- chitchat: Casual talk.
-
-History:
-{history}
-Examples:
-{formatted_examples}
-
-User: {user_input}
-Intent:"""
+) if embedding_model else None
+retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 6}) if vector_store else None
 
-def detect_intent(message: str) -> str:
-    prompt = build_intent_prompt(message, INTENT_EXAMPLES)
-    return llm.invoke(prompt).content.strip().lower()
 
 # Reformulation Prompt
 reformulate_prompt = PromptTemplate.from_template(
     """
-You are the main component of existence of a rag chat-bot you take in the last question and answer and a follow up question
-Basis the context of conversation and the follow up question, rewrite the question in a way that is more suite for the retreiver
-to search in documents 
+You are a smart medical assistant for a RAG chatbot.
+
+Tasks:
+1. Correct spelling mistakes in the input prompt, assuming a medical context.
+2. Rephrase any medical abbreviations to their full terms (e.g., IUFD to Intrauterine fetal death).
+3. Categorize and act:
+   - If only medical terminology, rephrase to "cure of [term]."
+   - If a new medical question, return as is.
+   - If a follow-up, rewrite based on last Q&A and context for document retrieval.
+   - If chitchat, return "chitchat."
 
 Previous Q&A:
 Q: {last_question}
@@ -136,9 +92,7 @@ def detect_intent(message: str) -> str:
 - Use context to address unanswered aspects from the previous answer.
 - Consider recent chat history for coherence.
 - Word limit: 150 words.
-
-After you've answered the query assess the context retreived,recent_history and what you answered
- and suggest relevant followup question like for eg would you like to know about...
+- Suggest a relevant follow-up question (e.g., "Would you like to know about...").
 
 Context:
 {context}
@@ -153,65 +107,32 @@ def detect_intent(message: str) -> str:
 )
 
 def route_intent(user_message: str):
-    # Preprocess abbreviations
-    for abbr, full_term in MEDICAL_ABBRS.items():
-        user_message = user_message.replace(abbr, full_term)
-    
-    intent = detect_intent(user_message)
+    #print("here 1")
+    if not llm or not retriever:
+        return "Error: System not fully initialized. Check dependencies."
+    #print("here 2")
     history = st.session_state.history
     last_turn = history[-1] if history else None
     prev_question, prev_answer, _ = last_turn if last_turn else ("", "", "")
     prev_chunks = st.session_state.last_chunks if st.session_state.last_chunks else []
-
-    if intent == "new_question":
-        docs = retriever.invoke(user_message)
+    
+    reform = reformulate_prompt.format(last_question=prev_question, last_answer=prev_answer, question=user_message)
+    rewritten = llm.invoke(reform).content.strip()
+    #print(rewritten)
+    if rewritten.lower() == "chitchat":
+        
+        answer= "I'm here for medical questions. Ask one!"
+    else:
+        docs = retriever.invoke(rewritten)
         chunks = [f"From [{os.path.basename(d.metadata.get('source', ''))}.pdf]: {d.page_content}" for d in docs]
         context = "\n\n".join(chunks[:3])  # Top 3 for brevity
-        prompt = answer_prompt_template.format(context=context, prev_answer="", prev_history="", question=user_message)
+        prompt = answer_prompt_template.format(context=context, prev_answer=prev_answer, prev_history=f"User: {prev_question} | Bot: {prev_answer}" if last_turn else "", question=rewritten)
         answer = llm.invoke(prompt).content
         st.session_state.last_chunks = docs
-
-    elif intent == "follow_up":
-        if last_turn and prev_chunks:
-            # Reformulate with prior context
-            reform = reformulate_prompt.format(last_question=prev_question, last_answer=prev_answer, question=user_message)
-            
-            rewritten = llm.invoke(reform).content
-            print(rewritten)
-            # Smart retrieval: Exclude prior chunk IDs, filter out None values
-            prev_ids = [doc.metadata.get('id') for doc in prev_chunks if doc.metadata.get('id') is not None]
-            new_docs = vector_store.similarity_search(rewritten, k=6, filter={"id": {"$nin": prev_ids}} if prev_ids else {})            
-            new_chunks = [f"From [{os.path.basename(d.metadata.get('source', ''))}.pdf]: {d.page_content}" for d in new_docs]
-            
-            # Combine context: 3 new + 1 past (if relevant)
-            combined_context = "\n\n".join(new_chunks[:3] + [prev_chunks[0].page_content] if prev_chunks else new_chunks[:3])
-            
-            # Prepare history
-            prev_history = f"User: {prev_question} | Bot: {prev_answer}"
-            
-            # Prompt with fused context
-            prompt = answer_prompt_template.format(
-                context=combined_context,
-                prev_answer=prev_answer,
-                prev_history=prev_history,
-                question=rewritten
-            )
-            answer = llm.invoke(prompt).content
-            st.session_state.last_chunks = new_docs
-        else:
-            answer = "Sorry, I need more context to answer."
-
-    elif intent == "gratitude":
-        answer = "You're welcome! Ask me anything else."
-
-    elif intent == "chitchat":
-        answer = "I'm here for medical questions. Ask one!"
-
-    else:
-        answer = "Sorry, I didn't understand. Please rephrase."
-
-    st.session_state.history.append((user_message, answer, intent))
+        #st.session_state.history.append((user_message, answer, "follow_up" if last_turn else "new_question"))
+    st.session_state.history.append((user_message, answer, "dummy"))
     return answer
+
 # Streamlit UI
 st.set_page_config(page_title="Medical Chatbot", layout="centered")
 st.title("🩺 Smart Medical Chatbot")
@@ -221,7 +142,7 @@ def route_intent(user_message: str):
 if user_input:
     with st.spinner("Thinking..."):
         response = route_intent(user_input)
-
+        #print(response)
 # Safely handle history display
 if st.session_state.history:
     for i, (q, a, intent) in enumerate(reversed(st.session_state.history)):