PDF Chatbot with LangChain and OpenAI

This repository contains a Streamlit application that allows users to chat with the content of an uploaded PDF file using OpenAI's GPT-4o. The application leverages LangChain for managing conversation flow and embedding PDFs into a vector database for efficient querying.

Features

• PDF Upload: Upload multi-page PDFs to the application.
• Vector Embedding: Parse and embed PDF content into a vector store using OpenAI embeddings.
• Chat Interface: Interact with the uploaded PDF content through a chat interface.
• State-of-the-Art Technologies: Uses the latest versions of Streamlit, LangChain, and OpenAI APIs.

Packages Used

• streamlit: For creating the web application interface.
• langchain: For managing the conversation flow and integrating with OpenAI.
• PyPDF2: For parsing PDF documents.
• FAISS: For efficient vector storage and similarity search.
• openai: For accessing OpenAI's GPT-4o API.

Directory Structure

ChatPDF/
├── app.py
├── requirements.txt
├── install_requirements.sh
├── sample_webapp.png
├── sample_demo.gif

Installation

1. Clone the repository:

   git clone https://github.com/xmpuspus/ChatPDF.git
   cd ChatPDF

2. Install the required packages:

   bash install_requirements.sh

3. Run the Streamlit application:

   streamlit run app.py

Application Overview

Sidebar Inputs

• OpenAI API Key: A text input for the user to enter their OpenAI API key.
• PDF Upload: A file uploader for users to upload their PDF documents.

Main Pane

• Chat Interface: Displays the chat history and allows users to input their questions.

Approach

1. PDF Upload and Parsing:

   • When a user uploads a PDF, it's saved temporarily and parsed using PyPDF2.
   • The extracted text from each page is concatenated into a single string.

2. Vector Embedding:

   • The parsed PDF text is embedded into a vector store using OpenAIEmbeddings from LangChain.
   • FAISS is used to store these embeddings and perform similarity searches efficiently.

3. Chat Handling:

   • User inputs are taken from the chat interface.
   • The most relevant text from the vector store is retrieved based on the user's query.
   • This relevant text, combined with the user's query, is used as a prompt for OpenAI's GPT-4o to generate a response.
   • The response is displayed in the chat interface.

User Experience

1. Upload PDF: Users upload their PDF documents through the sidebar.
2. Enter API Key: Users input their OpenAI API key to enable AI responses.
3. Chat Interaction: Users can ask questions related to the PDF content, and the AI will respond based on the parsed PDF text.

Security Considerations

• Injection-Proof: The application ensures that user inputs cannot override system instructions by strictly defining the prompt template and conversation flow.

Future Improvements

• Enhanced Parsing: Improve PDF parsing to handle complex documents better.
• Customization: Allow users to customize the prompt template and other settings.
• Multi-PDF Support: Enable handling of multiple PDFs for a broader range of queries.

Contributions

Contributions are welcome! Please open an issue or submit a pull request for any improvements or features you would like to see.

License

This project is licensed under the MIT License. See the LICENSE file for more details.