Multi-Document Conversational AI: Chat with PDFs, Word Docs, an…

Chinelo Nweke

AI Chatbot Developer

AI Developer

Python

Streamlit

Multi-Document Conversational AI: Chat with PDFs, Word Docs, and Text Files Using Streamlit.

Chinelo Nkem Nweke

7 min read

Jan 10, 2025

Introduction

In today’s world, being able to pull useful information from different types of documents — like PDFs, Word files (DOCX), spreadsheets (CSV), JSON files, and more — is super important for both businesses and researchers. This project is all about creating a chatbot app using Streamlit and LangChain. The cool part? This chatbot won’t just handle PDFs — it’ll work with various document and text file formats to make finding information quick and easy.

This app will also include a user login feature, ensuring each user can interact with their own data securely. The app won’t come with any preloaded data — users will upload their own documents, which will then be stored in the database. Once uploaded, users can ask the chatbot questions and get answers, but only based on their own data. This makes the app personalized, and tailored to each user’s specific needs. These are the steps I took in building and deploying the chatbot app:

Step 1: Environment and Requirement Setup

To kick things off, I started by creating a new repository on GitHub and cloning it to my local machine. This step is important because it gives me a clean, version-controlled space to work on the project. Once the repository was set up, I created an .env file to securely store my API keys. This way, I could keep credentials out of the code itself and make the app more secure.

Next, I created a requirements.txt file. This file lists all the necessary libraries and dependencies the project will need, such as streamlit, langchain, and others. By having this file, I can easily manage the project’s dependencies and ensure everything runs smoothly. This setup is pretty straightforward but essential because it ensures that the environment is ready to handle the rest of the app’s components.

import streamlit as stimport hashlibfrom dotenv import load_dotenvfrom PyPDF2 import PdfReaderfrom docx import Document from langchain.text_splitter import CharacterTextSplitterfrom langchain.embeddings import HuggingFaceEmbeddingsfrom langchain.vectorstores import FAISSfrom langchain_groq import ChatGroqfrom langchain.memory import ConversationBufferMemoryfrom langchain.chains import ConversationalRetrievalChainfrom htmlTemplates import css, bot_template, user_templatefrom langchain.llms import HuggingFaceHub

Step 2:User Authentication I set up user authentication so that users can register and log in using their username and password. To make sure their passwords stay secure, I use SHA-256 hashing, which protects their sensitive info. After logging in, each user can upload their own documents and interact with the chatbot, keeping their data private and accessible only to them.

# Hashing Function for Passwordsdef hash_password(password):    return hashlib.sha256(password.encode()).hexdigest()# Authentication Functionsdef register_user(username, password):    if "user_data" not in st.session_state:        st.session_state.user_data = {}            hashed_password = hash_password(password)    st.session_state.user_data[username] = hashed_password    st.success("Registration successful! Please log in.")def authenticate_user(username, password):    if "user_data" not in st.session_state:        st.session_state.user_data = {}    if not username or not password:        st.error("Username and password cannot be empty.")        return False    hashed_password = hash_password(password)    if username in st.session_state.user_data and st.session_state.user_data[username] == hashed_password:        return True    else:        st.error("Invalid username or password.")        return False

Step 3: Text Extraction from Documents Next, the app extracts text from the documents users upload. If they upload PDFs, I use the PyPDF2 library to pull the text from them. For Word documents (DOCX), the app uses the docx library to access the content. If the user uploads a text file, the app processes it just the same. This way, all the content from different file types is turned into raw text that the chatbot can work with.

# Function to extract text from PDF filesdef get_pdf_text(pdf_docs):    text = ""    for pdf in pdf_docs:        pdf_reader = PdfReader(pdf)        for page in pdf_reader.pages:            text += page.extract_text()    return text# Function to extract text from Word documents (.docx)def get_word_text(word_docs):    text = ""    for doc in word_docs:        doc = Document(doc)        for para in doc.paragraphs:            text += para.text + "\n"    return text# Function to extract text from text files (.txt)def get_text_file_text(text_files):    text = ""    for txt in text_files:        text += txt.read().decode("utf-8") + "\n"    return text

Step 4:Text Splitting Once the text is extracted, it needs to be broken down into smaller chunks. This is where LangChain’s CharacterTextSplitter comes in. This step is super important because it helps the chatbot search through long texts more easily. The text is split into manageable pieces, with a bit of overlap to make sure the context remains intact across different chunks.

# Function to split text into chunks for processingdef get_text_chunks(text):    text_splitter = CharacterTextSplitter(        separator="\n",        chunk_size=1000,        chunk_overlap=200,        length_function=len    )    chunks = text_splitter.split_text(text)    return chunks

Step 5:Vector Store Creation Then comes the part where the app turns the text into something the chatbot can really work with: embeddings. Using LangChain’s FAISS and HuggingFaceEmbeddings, the app converts those text chunks into numerical representations, which are stored in a vector database. This makes it quick and efficient to retrieve information when the user asks a question.

# Function to generate a vectorstore from text chunksdef get_vectorstore(text_chunks):    embeddings = HuggingFaceEmbeddings()    vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)    return vectorstore

Step 6: Building the Conversational Chain Now that we have the vector store with embeddings, the app sets up a conversation model using LangChain’s ChatGroq. The model comes with a memory buffer that keeps track of the chat history, allowing the chatbot to respond more intelligently by remembering what’s been said. This means users get answers based on the content they’ve uploaded.

# Function to get a conversational chain for the chat modeldef get_conversation_chain(vectorstore):    llm = ChatGroq()    memory = ConversationBufferMemory(        memory_key='chat_history', return_messages=True)    conversation_chain = ConversationalRetrievalChain.from_llm(        llm=llm,        retriever=vectorstore.as_retriever(),        memory=memory    )    return conversation_chain

Step 7: Handling User Input When a user asks a question, the app uses the conversation model to pull the most relevant info from the vector store. It updates the chat history and shows the conversation in real time. The app also refines the chatbot’s responses, cutting out unnecessary phrases and giving direct, clear answers to user queries.

# Process the conversation and display chat historydef handle_userinput(user_question):    response = st.session_state.conversation({'question': user_question})    st.session_state.chat_history = response['chat_history']        # Loop through the chat history and display messages    for i, message in enumerate(st.session_state.chat_history):        if i % 2 == 0:  # User's input            st.write(f"**🤖❓:** {message.content}")        else:  # Model's response            if message.content.lower().startswith("based on the information"):                answer = message.content[message.content.find('.')+1:].strip()                st.write(f"**🤖:** {answer}")            else:                st.write(f"**🤖:** {message.content}")

Step 8: User Interface and Document Upload The app is built using Streamlit, which makes the interface super simple and user-friendly. Users can register, log in, and upload documents directly through the sidebar. The file uploader handles multiple formats, and once the files are uploaded, the app processes them in the background, preparing everything for interaction. The chat interface makes it easy for users to ask questions and read the responses.

def main():    # Initialize session state variables    if "conversation" not in st.session_state:        st.session_state.conversation = None    if "chat_history" not in st.session_state:        st.session_state.chat_history = None    load_dotenv()    st.set_page_config(page_title="Chat with Documents", page_icon=":books:")    # Sidebar Login/Register    with st.sidebar:        st.subheader("Login/Register")        choice = st.radio("Select an option", ["Login", "Register"])        username = st.text_input("Username")        password = st.text_input("Password", type="password")        if choice == "Register":            if st.button("Register"):                if username and password:                    register_user(username, password)                else:                    st.warning("Please provide both username and password.")        elif choice == "Login":            if st.button("Login"):                if username and password:                    if authenticate_user(username, password):                        st.success("Login successful!")                        st.session_state.logged_in_user = username                    else:                        st.error("Invalid username or password.")                else:                    st.warning("Please provide both username and password.")    # Main app after login    if "logged_in_user" in st.session_state:        st.header(f"Welcome, {st.session_state.logged_in_user}!")        user_question = st.text_input("Ask a question about your documents:")        if user_question:            handle_userinput(user_question)        with st.sidebar:            st.subheader("Your Documents")            uploaded_files = st.file_uploader(                "Upload your PDFs, Word docs, and text files here", accept_multiple_files=True            )            if st.button("Process"):                with st.spinner("Processing..."):                    raw_text = ""                    pdf_files = [file for file in uploaded_files if file.type == "application/pdf"]                    word_files = [file for file in uploaded_files if file.type == "application/vnd.openxmlformats-officedocument.wordprocessingml.document"]                    text_files = [file for file in uploaded_files if file.type == "text/plain"]                    if pdf_files:                        raw_text += get_pdf_text(pdf_files)                    if word_files:                        raw_text += get_word_text(word_files)                    if text_files:                        raw_text += get_text_file_text(text_files)                    text_chunks = get_text_chunks(raw_text)                    vectorstore = get_vectorstore(text_chunks)                    st.session_state.conversation = get_conversation_chain(vectorstore)if __name__ == "__main__":    main()

Step 9: Running the App Finally, the app is designed to run smoothly whether locally or on a cloud platform like Streamlit Cloud. The main function kicks things off by initializing all the necessary components — from authentication to document processing and user interactions. Once everything is set up, users are ready to start interacting with their documents in a personalized, engaging way.

Conclusion:

By combining all these tools and technologies, the app makes it easy for users to extract valuable insights from their documents in a seamless, scalable way. Whether it’s PDFs, Word docs, or text files, the chatbot can help users find exactly what they’re looking for.

Note: This is a personal project but for a business, the app should have the ability to store data from user chat history.

You can interact with the app via this link

Here is link to the Github repository

Thank you for reading and please leave your opinion.

You can find me here;

LinkedIn: Chinelo Nweke

x: Nelo Nweke

Like this project

Posted Jan 30, 2025

In today’s world, being able to pull useful information from different types of documents — like PDFs, Word files (DOCX), spreadsheets (CSV), JSON files, and m…

Likes

Views