Upload 3 files

app.py

@@ -0,0 +1,384 @@
+import streamlit as st
+import streamlit_chat
+import os
+from config import app_name
+from config import website_name
+from config import DATABASE
+from config import PINECONE_INDEX
+from config import CHAT_COLLECTION
+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
+import json
+
+st.set_page_config(layout="wide", page_title=app_name, 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)
+FLASH_API = os.getenv("FLASH_API")
+PINECONE_API = os.getenv("PINECONE_API_KEY")
+MONGO_URI = os.getenv("MONGO_URI")
+
+pc = pinecone.Pinecone(
+    api_key=PINECONE_API
+)
+
+index = pc.Index(PINECONE_INDEX)
+# MongoDB connection setup
+
+client = MongoClient(MONGO_URI)
+db = client[DATABASE]
+chat_sessions = db[CHAT_COLLECTION]
+
+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, language):
+    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",
+             f"""You are a website-specific chatbot specializing in answering user queries about {website_name}. You will be provided with data chunks sourced from {website_name}, and each chunk has an associated URL and the language in which you have to respond. When formulating your responses:
+               1. Clarity and Completeness
+                  - Always strive to deliver thorough, concise, and direct answers.
+                  - If the user’s query is ambiguous or there are multiple possible answers, ask for clarification with a clear rationale.
+               2. No Chunk Names
+                  - Do not reference chunk filenames or mention the term “chunk” in your replies.
+                  - Instead, present the information in a natural, conversational style.
+
+               3. Use of Conversation History
+                  - Refer back to  conversation history for consistency and to get context for a follow up question.
+                  - If there are previous statements like “The answer is not available,” ignore them unless still relevant to the current query.
+
+               4. Handling Off-Topic Queries
+                  - If the user sends greetings, introductions, or queries unrelated to {website_name}, respond politely and conversationally without forcing a website-related answer.
+
+               5. Source URLs
+                  - Always provide the URLs you used to answer the query under a “Sources” heading at the end of your reply.
+                  - If the same URL appears in multiple relevant chunks, list that URL only once in the sources section.
+                  - Only include the URLs that genuinely informed or supported your answer.
+                  - if the answer itself contains url, then quote it properly
+
+               6. No Direct Chunk Quotes
+                  - Summarize or paraphrase the original content instead of quoting chunk names or raw text verbatim (unless absolutely necessary for clarity).
+               7. Relevance Check 
+                  - Thoroughly check the provided data chunks before replying.
+                  - If you can’t find an answer in the chunks, or if the query is irrelevant politely ask for clarification or explain that you cannot answer.
+                  
+               8. Response Language
+                  - You will be provided the language in which you must answer.
+                  - Note that the chunks will be in english, so you will have to translate them and then respond.
+
+               8. Formatting
+                  - Present your answers in a well-structured format—either bullet points or clear paragraphs—to ensure maximum readability.
+
+
+                            """),
+
+            ("human", f'''
+                            "Query":\n {query}\n
+                            "Language": {language}
+                            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, "Language":language, "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", f""""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.
+                           If the user query is not in english and some other Indian language such as Hindi, Gujarati, Telugu, Tamil, Kannada etc. your task is to translate it to  english and then generate the improved query in english.
+                           The provided queries are related to {website_name}.
+                           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.
+                           
+                           Output the result in JSON
+                           expected_output_format:
+                           "query" : "String or None",
+                           "original_query_language" : "String"  #identify the original query language
+                           
+                           """),
+            ("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):
+    cleaned_response = re.sub(r"^`*json\s*", "", llm_output)
+    cleaned_response = re.sub(r"`*$", "", cleaned_response)
+    if cleaned_response:
+        try:
+            json_data = json.loads(cleaned_response)
+            return json_data
+        except json.JSONDecodeError as e:
+            print(f"JSON decoding error: {e}")
+            return None
+
+    else:
+        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(f'<div class="fixed-header"><h1>Welcome To {website_name}  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 {website_name}")
+
+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_json = get_query_from_llm_answer(query)
+            extracted_query = extracted_query_json.get("query")
+            language = extracted_query_json.get("original_query_language")
+            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'],language)
+
+            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()

config.py

@@ -0,0 +1,6 @@
+app_name="PM Kisan Samman Nidhi Chatbot"
+website_name="PM Kisan Samman Nidhi"
+PINECONE_INDEX="kisan"
+DATABASE="chatbots"
+CHUNK_COLLECTION="kisan_chunks"
+CHAT_COLLECTION="kisan_pfrda"

requirements.txt

@@ -0,0 +1,113 @@
+aiohappyeyeballs==2.4.4
+aiohttp==3.11.11
+aiosignal==1.3.2
+altair==5.5.0
+annotated-types==0.7.0
+anyio==4.7.0
+async-timeout==4.0.3
+attrs==24.3.0
+beautifulsoup4==4.12.3
+blinker==1.9.0
+cachetools==5.5.0
+certifi==2024.12.14
+charset-normalizer==3.4.1
+click==8.1.8
+colorama==0.4.6
+dataclasses-json==0.6.7
+distro==1.9.0
+dnspython==2.7.0
+exceptiongroup==1.2.2
+filetype==1.2.0
+frozenlist==1.5.0
+fuzzywuzzy==0.18.0
+gitdb==4.0.11
+GitPython==3.1.43
+google-ai-generativelanguage==0.6.10
+google-api-core==2.24.0
+google-api-python-client==2.156.0
+google-auth==2.37.0
+google-auth-httplib2==0.2.0
+google-generativeai==0.8.3
+googleapis-common-protos==1.66.0
+greenlet==3.1.1
+grpcio==1.68.1
+grpcio-status==1.68.1
+h11==0.14.0
+httpcore==1.0.7
+httplib2==0.22.0
+httpx==0.28.1
+httpx-sse==0.4.0
+idna==3.10
+Jinja2==3.1.5
+jiter==0.8.2
+jsonpatch==1.33
+jsonpointer==3.0.0
+jsonschema==4.23.0
+jsonschema-specifications==2024.10.1
+langchain==0.3.13
+langchain-community==0.3.13
+langchain-core==0.3.28
+langchain-google-genai==2.0.7
+langchain-text-splitters==0.3.4
+langsmith==0.2.6
+Levenshtein==0.26.1
+markdown-it-py==3.0.0
+markdownify==0.14.1
+MarkupSafe==3.0.2
+marshmallow==3.23.2
+mdurl==0.1.2
+multidict==6.1.0
+mypy-extensions==1.0.0
+narwhals==1.19.1
+numpy==1.26.4
+openai==1.58.1
+orjson==3.10.12
+packaging==24.2
+pandas==2.2.3
+pillow==11.0.0
+pinecone==5.4.2
+pinecone-plugin-inference==3.1.0
+pinecone-plugin-interface==0.0.7
+propcache==0.2.1
+proto-plus==1.25.0
+protobuf==5.29.2
+pyarrow==18.1.0
+pyasn1==0.6.1
+pyasn1_modules==0.4.1
+pydantic==2.10.4
+pydantic-settings==2.7.0
+pydantic_core==2.27.2
+pydeck==0.9.1
+Pygments==2.18.0
+pymongo==4.10.1
+pyparsing==3.2.0
+python-dateutil==2.9.0.post0
+python-dotenv==1.0.1
+python-Levenshtein==0.26.1
+pytz==2024.2
+PyYAML==6.0.2
+RapidFuzz==3.11.0
+referencing==0.35.1
+requests==2.32.3
+requests-toolbelt==1.0.0
+rich==13.9.4
+rpds-py==0.22.3
+rsa==4.9
+six==1.17.0
+smmap==5.0.1
+sniffio==1.3.1
+soupsieve==2.6
+SQLAlchemy==2.0.36
+streamlit==1.41.1
+streamlit-chat==0.1.1
+tenacity==9.0.0
+toml==0.10.2
+tornado==6.4.2
+tqdm==4.67.1
+typing-inspect==0.9.0
+typing_extensions==4.12.2
+tzdata==2024.2
+uritemplate==4.1.1
+urllib3==2.3.0
+watchdog==6.0.0
+yarl==1.18.3