app.py

import streamlit as st
from langchain_openai import ChatOpenAI
import os
import dotenv
from langchain_community.document_loaders import WebBaseLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import HumanMessage, AIMessage
from langchain.memory import ConversationBufferMemory
from langchain.document_loaders import PyPDFLoader

# Set page config
st.set_page_config(page_title="Tbank Assistant", layout="wide")

# Streamlit app header
st.title("Tbank Customer Support Chatbot")

# Sidebar for API Key input
with st.sidebar:
 st.header("Configuration")
 api_key = st.text_input("Enter your OpenAI API Key:", type="password")
 if api_key:
 os.environ["OPENAI_API_KEY"] = api_key

# Main app logic
if "OPENAI_API_KEY" in os.environ:
 # Initialize components
 @st.cache_resource
 def initialize_components():
 dotenv.load_dotenv()
 chat = ChatOpenAI(model="gpt-3.5-turbo-1106", temperature=0.2)

 #loader1 = WebBaseLoader("https://www.tbankltd.com/")
 loader1 = PyPDFLoader("Tbank resources.pdf")
 loader2 = PyPDFLoader("International Banking Services.pdf")
 data1 = loader1.load()
 data2 = loader2.load()
 data = data1 + data2
 text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
 all_splits = text_splitter.split_documents(data)
 vectorstore = Chroma.from_documents(documents=all_splits, embedding=OpenAIEmbeddings())
 retriever = vectorstore.as_retriever(k=4)

 SYSTEM_TEMPLATE = """
 You are Tbank's AI assistant, a chatbot whose knowledge comes exclusively from Tbank's provided PDF documents. Follow these guidelines:

 Guidelines:
 Identity Confirmation:
 If asked, state: "I am Tbank's AI assistant. How can I help you today?"
 Scope of Information:
 Use only information from Tbank's website content and provided PDF documents.
 Do not infer or provide information from outside these sources.
 Response Style:
 Provide clear, concise responses.
 Keep answers brief and relevant to the user's query.
 Maintain a friendly and professional tone.
 Unknown Information:
 If a query is outside your knowledge base, respond: "I apologize, but I don't have information about that. My knowledge is limited to Tbank's products/services and our website/document content. Is there anything specific about Tbank I can help with?"
 If unsure about an answer, say: "I'm not certain about that. For accurate information, please check our website or contact our customer support team."
 Factual Information:
 Remind users that you provide only factual information from Tbank sources.
 End Interaction:
 Always end by asking: "Is there anything else you can help with regarding Tbank?"
 Examples:
 General Greeting:
 
 "Hello! Welcome to Tbank. How can I assist you today?"
 Identity Query:
 
 "I am Tbank's AI assistant. How can I help you today?"
 Out of Scope Query:
 
 "I apologize, but I don't have information about that. My knowledge is limited to Tbank's products/services and our website/document content. Is there anything specific about Tbank I can help with?"
 Uncertainty:
 
 "I'm not certain about that. For accurate information, please check our website or contact our customer support team."
 Closing:
 
 "Is there anything else you can help with regarding Tbank?"

 <context>
 {context}
 </context>

 Chat History:
 {chat_history}
 """

 question_answering_prompt = ChatPromptTemplate.from_messages(
 [
 (
 "system",
 SYSTEM_TEMPLATE,
 ),
 MessagesPlaceholder(variable_name="chat_history"),
 MessagesPlaceholder(variable_name="messages"),
 ]
 )

 document_chain = create_stuff_documents_chain(chat, question_answering_prompt)

 return retriever, document_chain

 # Load components
 with st.spinner("Initializing Tbank Assistant..."):
 retriever, document_chain = initialize_components()

 # Initialize memory for each session
 if "memory" not in st.session_state:
 st.session_state.memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

 # Chat interface
 st.subheader("Chat with Tbank Assistant")

 # Initialize chat history
 if "messages" not in st.session_state:
 st.session_state.messages = []

 # Display chat messages from history on app rerun
 for message in st.session_state.messages:
 with st.chat_message(message["role"]):
 st.markdown(message["content"])

 # React to user input
 if prompt := st.chat_input("What would you like to know about Tbank?"):
 # Display user message in chat message container
 st.chat_message("user").markdown(prompt)
 # Add user message to chat history
 st.session_state.messages.append({"role": "user", "content": prompt})

 with st.chat_message("assistant"):
 message_placeholder = st.empty()

 # Retrieve relevant documents
 docs = retriever.get_relevant_documents(prompt)

 # Generate response
 response = document_chain.invoke(
 {
 "context": docs,
 "chat_history": st.session_state.memory.load_memory_variables({})["chat_history"],
 "messages": [
 HumanMessage(content=prompt)
 ],
 }
 )

 # The response is already a string, so we can use it directly
 full_response = response
 message_placeholder.markdown(full_response)

 # Add assistant response to chat history
 st.session_state.messages.append({"role": "assistant", "content": full_response})

 # Update memory
 st.session_state.memory.save_context({"input": prompt}, {"output": full_response})

else:
 st.warning("Please enter your OpenAI API Key in the sidebar to start the chatbot.")

# Add a footer
st.markdown("---")
st.markdown("By AI Planet")