Upload 2 files

app.py

@@ -0,0 +1,338 @@
+import streamlit as st
+import streamlit_chat
+import json
+import os
+from pymongo import MongoClient
+from bson import ObjectId
+from dotenv import load_dotenv
+import pinecone
+from langchain_google_genai import GoogleGenerativeAIEmbeddings
+from langchain_google_genai import ChatGoogleGenerativeAI
+from langchain_core.prompts import ChatPromptTemplate
+import re
+
+st.set_page_config(layout="wide", page_title="ESIC Chatbot", page_icon="📄")
+load_dotenv()
+import logging
+from pytz import timezone, utc
+from datetime import datetime
+
+logging.basicConfig(
+    level=logging.DEBUG,  # This is for your application logs
+    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
+    datefmt='%Y-%m-%d %H:%M:%S'
+)
+
+# Suppress pymongo debug logs by setting the pymongo logger to a higher level
+pymongo_logger = logging.getLogger('pymongo')
+pymongo_logger.setLevel(logging.WARNING)
+
+PINECONE_API = os.getenv("PINECONE_API_KEY")
+pc = pinecone.Pinecone(
+    api_key=PINECONE_API
+)
+index_name = "esic"
+index = pc.Index(index_name)
+# MongoDB connection setup
+MONGO_URI = os.getenv("MONGO_URI")
+client = MongoClient(MONGO_URI)
+db = client["esic"]
+chat_sessions = db["chat_session"]
+
+# Set LLM models
+FLASH_API = os.getenv("FLASH_API")
+embeddings = GoogleGenerativeAIEmbeddings(model="models/embedding-001", google_api_key=FLASH_API)
+llm = ChatGoogleGenerativeAI(model="gemini-1.5-flash", temperature=0, max_tokens=None, google_api_key=FLASH_API)
+
+# Load the extracted JSON data
+
+# Initialize session state for current chat session
+if 'current_chat_id' not in st.session_state:
+    st.session_state['current_chat_id'] = None
+if 'chat_history' not in st.session_state:
+    st.session_state['chat_history'] = []
+if 'regenerate' not in st.session_state:
+    st.session_state['regenerate'] = False  # Track regenerate button state
+
+
+# Function to create a new chat session in MongoDB
+def create_new_chat_session():
+    # Get the current time in IST
+    ind_time = datetime.now(timezone("Asia/Kolkata"))
+    # Convert IST time to UTC for storing in MongoDB
+    utc_time = ind_time.astimezone(utc)
+
+    new_session = {
+        "created_at": utc_time,  # Store in UTC
+        "messages": []  # Empty at first
+    }
+    session_id = chat_sessions.insert_one(new_session).inserted_id
+    return str(session_id)
+
+
+# Function to load a chat session by MongoDB ID
+# Function to load the chat session by MongoDB ID (load full history for display)
+def load_chat_session(session_id):
+    session = chat_sessions.find_one({"_id": ObjectId(session_id)})
+    if session:
+        # Load the full chat history (no slicing here)
+        st.session_state['chat_history'] = session['messages']
+
+
+# Function to update chat session in MongoDB (store last 15 question-answer pairs)
+# Function to update chat session in MongoDB (store entire chat history)
+def update_chat_session(session_id, question, answer, improved_question):
+    # Append the new question-answer pair to the full messages array
+    chat_sessions.update_one(
+        {"_id": ObjectId(session_id)},
+        {"$push": {
+            "messages": {"$each": [{"question": question, 'improved_question': improved_question, "answer": answer}]}}}
+    )
+
+
+# Function to replace the last response in MongoDB
+def replace_last_response_in_mongo(session_id, new_answer):
+    last_message_index = len(st.session_state['chat_history']) - 1
+    if last_message_index >= 0:
+        # Replace the last response in MongoDB
+        chat_sessions.update_one(
+            {"_id": ObjectId(session_id)},
+            {"$set": {f"messages.{last_message_index}.answer": new_answer}}
+        )
+
+
+# Function to regenerate the response
+def regenerate_response():
+    try:
+        if st.session_state['chat_history']:
+            last_question = st.session_state['chat_history'][-1]["question"]  # Get the last question
+            # Exclude the last response from the history when sending the question to LLM
+            previous_history = st.session_state['chat_history'][:-1]  # Exclude the last Q&A pair
+
+            with st.spinner("Please wait, regenerating the response!"):
+                # Generate a new response for the last question using only the previous history
+
+                query = get_context_from_messages(last_question, previous_history)
+                if query:
+                    logging.info(f"Extracted query is :{query}\n")
+                    extracted_query = get_query_from_llm_answer(query)
+                    if extracted_query:
+                        query = extracted_query
+                    else:
+                        query = last_question
+
+                    query_embedding = embeddings.embed_query(query)
+                    search_results = index.query(vector=query_embedding, top_k=10, include_metadata=True)
+                    matches = search_results['matches']
+
+                    content = ""
+                    for i, match in enumerate(matches):
+                        chunk = match['metadata']['chunk']
+                        url = match['metadata']['url']
+                        content += f"chunk{i}: {chunk}\n" + f"url{i}: {url}\n"
+
+                    new_reply = generate_summary(content, query, previous_history)
+
+                    st.session_state['chat_history'][-1]["answer"] = new_reply
+
+                    # Update MongoDB with the new response
+                    if st.session_state['current_chat_id']:
+                        replace_last_response_in_mongo(st.session_state['current_chat_id'], new_reply)
+
+                    st.session_state['regenerate'] = False  # Reset regenerate flag
+                    st.rerun()
+
+    except Exception as e:
+        st.error("Error occured in Regenerating response, please try again later.")
+
+
+def generate_summary(chunks, query, chat_history):
+    try:
+        # Limit the history sent to the LLM to the latest 3 question-answer pairs
+        limited_history = chat_history[-3:] if len(chat_history) > 3 else chat_history
+
+        # Create conversation history for the LLM, only using the last 15 entries
+        history_text = "\n".join([f"User: {q['improved_question']}\nLLM: {q['answer']}" for q in limited_history])
+
+        # Define the system and user prompts including the limited history
+        prompt = ChatPromptTemplate.from_messages([
+            ("system", """You are a chatbot specializing in answering queries related to Employees State Insurance Corporation(ESIC) website. You will be provided with chunks of data from the ESIC website to answer user queries. Each chunk will include associated URLs, You must give the url of the chunks which you are using to answer the query.
+                            Key Guidelines:
+                            1.If the user query is not clear, or you think multiple answers are possbile, you should ask for clarification with proper reasoning.
+                            2.Do not mention chunk name in any of your replies.
+                            3.Detailed and Clear: Provide thorough, clear, and concise responses without omitting relevant information from the data chunks.
+                            4.Natural Summarization: When answering, you must not  directly quote chunk names,formats. Instead, summarize or interpret the data naturally and conversationally.
+                            5.Use Conversation History: Refer back to the conversation history to maintain consistency and build on prior responses, if applicable.
+                            6.Ignore Unanswered Queries: If the conversation history contains previous responses like "The answer is not available in the context," disregard them when formulating your current response.
+                            7.Graceful Handling of General Queries: If a user sends greetings, introduction, salutations, or unrelated questions, respond appropriately and conversationally.
+                            8.Include Source URLs: Always include the URLs from the relevant chunks of data that you're using to answer the query.
+                            9.Thoroughly looks for answer to the query in the provided chunks before replying, if you feel the query is irrelevant or answer is not present then you can ask user to clarify or tell that it cannot be answered.
+                            10.Sometimes chunks might contain very less data still use it if its relevant.
+                            11. Its important to give the answer in a well formatted structure, like points or paragraphs
+                            """),
+
+            ("human", f'''
+                            "Query":\n {query}\n
+                            Below are the pinecone chunks that should be used to answer the user query:
+                            "Extracted Data": \n{chunks}\n
+                            Below is the previous conversation history:
+                            "Previous Conversation History": \n{history_text}\n
+                            '''
+             )
+        ])
+
+        # Chain the prompt with LLM for response generation
+        chain = prompt | llm
+        result = chain.invoke({"Query": query, "Extracted Data": chunks, "Previous Conversation History": history_text})
+
+        # Return the generated response
+        logging.info(f"LLM answer is :{result}")
+        return result.content
+
+    except Exception as e:
+        st.error(f"Error answering your question: {e}")
+        return None
+
+
+def get_context_from_messages(query, chat_history):
+    try:
+
+        logging.info(f"Getting context from original query: {query}")
+
+        # Limit the history sent to the LLM to the latest 3 question-answer pairs
+        limited_history = chat_history[-3:] if len(chat_history) > 3 else chat_history
+
+        # Create conversation history for the LLM, only using the last 15 entries
+        history_text = "\n".join([f"User: {q['question']}\nLLM: {q['answer']}" for q in limited_history])
+
+        # Define the system and user prompts including the limited history
+        prompt = ChatPromptTemplate.from_messages([
+            ("system", """"I will provide you with a user query and up to the last 3 messages from the chat history which includes both questions and answers.Your task is to understand  the user query nicely and restructure it if required such that  it  makes complete sense and is completely self contained.
+                           The provided queries are related to Employees State Insurance Corporation(ESIC).
+                           1. If the query is a follow-up, use the provided chat history to reconstruct a well-defined, contextually complete query that can stand alone."
+                           2. if the query is self contained, if applicable try to improve it to make is coherent.
+                           3. if the user query is salutations, greetings or not relevant in that case give the query back as it is.
+                           I have provided an output format below, stricly follow it. Do not give anything else other than just the output.
+                           expected_output_format: "query: String or None"
+                           """),
+            ("human", f'''  
+                            "Query":\n {query}\n
+                            "Previous Conversation History": \n{history_text}\n
+                            '''
+             )
+        ])
+
+        # Chain the prompt with LLM for response generation
+        chain = prompt | llm
+        result = chain.invoke({"Query": query, "Previous Conversation History": history_text})
+        logging.info(f"llm answer for query extraction is :{result}")
+
+        # Return the generated response
+        return result.content
+
+    except Exception as e:
+        logging.error(f"exception occured in getting query from original query :{e}")
+        return None
+
+
+def get_query_from_llm_answer(llm_output):
+    match = re.search(r'query:\s*(.*)', llm_output)
+    if match:
+        query = match.group(1).strip().strip('"')  # Remove leading/trailing spaces and quotes
+        return None if query.lower() == "none" else query
+    return None
+
+
+# Sidebar for showing chat sessions and creating new sessions
+st.sidebar.header("Chat Sessions")
+
+# Button for creating a new chat
+if st.sidebar.button("New Chat"):
+    new_chat_id = create_new_chat_session()
+    st.session_state['current_chat_id'] = new_chat_id
+    st.session_state['chat_history'] = []
+
+# List existing chat sessions with delete button (dustbin icon)
+existing_sessions = chat_sessions.find().sort("created_at", -1)
+for session in existing_sessions:
+    session_id = str(session['_id'])
+
+    # Retrieve stored UTC time and convert it to IST for display
+    utc_time = session['created_at']
+    ist_time = utc_time.replace(tzinfo=utc).astimezone(timezone("Asia/Kolkata"))
+    session_date = ist_time.strftime("%Y-%m-%d %H:%M:%S")  # Format for display
+
+    col1, col2 = st.sidebar.columns([8, 1])
+    with col1:
+        if st.button(f"Session {session_date}", key=session_id):
+            st.session_state['current_chat_id'] = session_id
+            load_chat_session(session_id)
+
+    # Display delete icon (dustbin)
+    with col2:
+        if st.button("🗑️", key=f"delete_{session_id}"):
+            chat_sessions.delete_one({"_id": ObjectId(session_id)})
+            st.rerun()  # Refresh the app to remove the deleted session from the sidebar
+
+# Main chat interface
+st.markdown('<div class="fixed-header"><h1>Welcome To ESIC Chatbot</h1></div>', unsafe_allow_html=True)
+st.markdown("<hr>", unsafe_allow_html=True)
+
+# Input box for the question
+user_question = st.chat_input(f"Ask a Question related to ESIC Website")
+
+if user_question:
+    # Automatically create a new session if none exists
+    if not st.session_state['current_chat_id']:
+        new_chat_id = create_new_chat_session()
+        st.session_state['current_chat_id'] = new_chat_id
+
+    with st.spinner("Please wait, I am thinking!!"):
+        # Store the user's question and get the assistant's response
+        query = get_context_from_messages(user_question, st.session_state['chat_history'])
+        if query:
+            logging.info(f"Extracted query is :{query}\n")
+            extracted_query = get_query_from_llm_answer(query)
+            if extracted_query:
+                query = extracted_query
+            else:
+                query = user_question
+
+            query_embedding = embeddings.embed_query(query)
+            search_results = index.query(vector=query_embedding, top_k=10, include_metadata=True)
+            matches = search_results['matches']
+
+            content = ""
+            for i, match in enumerate(matches):
+                chunk = match['metadata']['chunk']
+                url = match['metadata']['url']
+                content += f"chunk{i}: {chunk}\n" + f"url{i}: {url}\n"
+
+            print(f"content being passed is {content}")
+            reply = generate_summary(content, query, st.session_state['chat_history'])
+
+            if reply:
+                # Append the new question-answer pair to chat history
+                st.session_state['chat_history'].append(
+                    {"question": user_question, "answer": reply, "improved_question": query})
+
+                # Update the current chat session in MongoDB
+                if st.session_state['current_chat_id']:
+                    update_chat_session(st.session_state['current_chat_id'], user_question, reply, query)
+
+            else:
+                st.error("Error processing your request, Please try again later.")
+        else:
+            st.error("Error processing your request, Please try again later.")
+# Display the updated chat history (show last 15 question-answer pairs)
+for i, pair in enumerate(st.session_state['chat_history']):
+    question = pair["question"]
+    answer = pair["answer"]
+    streamlit_chat.message(question, is_user=True, key=f"chat_message_user_{i}")
+    streamlit_chat.message(answer, is_user=False, key=f"chat_message_assistant_{i}")
+
+# Display regenerate button under the last response
+if st.session_state['chat_history'] and not st.session_state['regenerate']:
+    if st.button("🔄 Regenerate", key="regenerate_button"):
+        st.session_state['regenerate'] = True
+        regenerate_response()

requirements.txt

@@ -0,0 +1,102 @@
+aiohappyeyeballs==2.4.0
+aiohttp==3.10.5
+aiosignal==1.3.1
+altair==5.4.1
+annotated-types==0.7.0
+anyio==4.4.0
+async-timeout==4.0.3
+attrs==24.2.0
+blinker==1.8.2
+boto3==1.35.15
+botocore==1.35.15
+cachetools==5.5.0
+certifi==2024.8.30
+charset-normalizer==3.3.2
+click==8.1.7
+colorama==0.4.6
+dnspython==2.6.1
+exceptiongroup==1.2.2
+frozenlist==1.4.1
+gitdb==4.0.11
+GitPython==3.1.43
+google-ai-generativelanguage==0.6.6
+google-api-core==2.19.2
+google-api-python-client==2.144.0
+google-auth==2.34.0
+google-auth-httplib2==0.2.0
+google-generativeai==0.7.2
+googleapis-common-protos==1.65.0
+greenlet==3.1.0
+grpcio==1.66.1
+grpcio-status==1.62.3
+h11==0.14.0
+httpcore==1.0.5
+httplib2==0.22.0
+httpx==0.27.2
+idna==3.8
+Jinja2==3.1.4
+jmespath==1.0.1
+jsonpatch==1.33
+jsonpointer==3.0.0
+jsonschema==4.23.0
+jsonschema-specifications==2023.12.1
+langchain==0.2.16
+langchain-core==0.2.38
+langchain-google-genai==1.0.10
+langchain-text-splitters==0.2.4
+langsmith==0.1.117
+markdown-it-py==3.0.0
+MarkupSafe==2.1.5
+mdurl==0.1.2
+multidict==6.1.0
+narwhals==1.6.4
+numpy==1.26.4
+orjson==3.10.7
+packaging==24.1
+pandas==2.2.2
+pillow==10.4.0
+pinecone==5.1.0
+pinecone-client==5.0.1
+pinecone-plugin-inference==1.0.3
+pinecone-plugin-interface==0.0.7
+proto-plus==1.24.0
+protobuf==4.25.4
+pyarrow==17.0.0
+pyasn1==0.6.0
+pyasn1_modules==0.4.0
+pydantic==2.9.1
+pydantic_core==2.23.3
+pydeck==0.9.1
+Pygments==2.18.0
+pymongo==4.8.0
+pyparsing==3.1.4
+PyPDF2==3.0.1
+python-dateutil==2.9.0.post0
+python-dotenv==1.0.1
+pytz==2024.1
+PyYAML==6.0.2
+referencing==0.35.1
+requests==2.32.3
+rich==13.8.1
+router==0.1
+rpds-py==0.20.0
+rsa==4.9
+s3transfer==0.10.2
+six==1.16.0
+smmap==5.0.1
+sniffio==1.3.1
+SQLAlchemy==2.0.34
+streamlit==1.38.0
+streamlit-chat==0.1.1
+streamlit-router==0.1.8
+tenacity==8.5.0
+toml==0.10.2
+tornado==6.4.1
+tqdm==4.66.5
+typing_extensions==4.12.2
+tzdata==2024.1
+uritemplate==4.1.1
+urllib3==2.2.2
+watchdog==4.0.2
+Werkzeug==3.0.4
+yarl==1.11.1