PDFGPT

Overview

PDFGPT is an advanced PDF Question Answering (QA) system that leverages Natural Language Processing (NLP) and machine learning models to provide intelligent responses to queries about PDF content. This project combines a FastAPI backend server with a React frontend client, all containerized using Docker for seamless deployment and scalability.

The system processes uploaded PDFs, extracts text, and utilizes cutting-edge NLP techniques to deliver accurate and context-aware answers to user queries. By integrating powerful language models and semantic search capabilities, PDFGPT offers an efficient solution for extracting information from lengthy documents.

This project was inspired by and aims to improve upon the functionality of https://github.com/bhaskatripathi/pdfGPT. Due to operational issues with the original implementation, PDFGPT was developed as a more robust and reliable alternative, incorporating additional features and optimizations to enhance performance and user experience.

Screenshot

Below is a screenshot of PDFGPT in action:

Purpose and Description

PDFGPT is designed to revolutionize the way users interact with PDF documents. By allowing users to upload PDFs and ask questions about their content, PDFGPT eliminates the need for manual searching through lengthy documents. The system processes PDFs, extracts text, and utilizes cutting-edge NLP techniques to deliver accurate and context-aware answers to user queries.

Key Features

Effortless PDF upload and processing
Multi-file support: Ability to upload and process multiple PDF files simultaneously, allowing users to query across multiple documents.
Deferred text embedding: Upload PDFs first, then perform text embedding for all uploaded files when ready.
Advanced text extraction and embedding
Semantic search for pinpointing relevant content
AI-powered question answering using state-of-the-art language models
Intuitive web interface for seamless user interaction
Choice between two text embedding models: Universal Sentence Encoder (USE) and OpenAI's Ada v2 (ADA)
Bilingual answer generation: Option to receive answers in English or Korean

PDFGPT is invaluable across various domains including research, education, legal, and information retrieval, where quick access to specific information within extensive documents is crucial.

Technology Stack

PDFGPT utilizes a modern and robust technology stack to provide efficient PDF processing and question answering capabilities.

Backend (Server)

Framework: FastAPI - A modern, fast (high-performance) web framework for building APIs with Python 3.7+ based on standard Python type hints.
Language: Python 3.11+
Key Libraries and Models:
- tensorflow and tensorflow-hub for machine learning and NLP tasks
- Universal Sentence Encoder (Version 4) - A pre-trained model from TensorFlow Hub used for generating sentence embeddings.
- OpenAI Ada v2 - A text embedding model from OpenAI that converts text into high-dimensional vectors, enabling semantic search and improving QA accuracy.
- PyMuPDF for PDF processing
- scikit-learn for machine learning utilities
- numpy for numerical computations
- openai for OpenAI API integration to leverage advanced language models
API Documentation: Swagger UI (automatically generated by FastAPI)
Containerization: Docker

Frontend (Client)

Framework: React - A JavaScript library for building user interfaces
Language: JavaScript/TypeScript
Key Libraries:
- axios for HTTP requests
- (If used, mention other significant libraries like react-router, redux, etc.)
Containerization: Docker
Web Server: Nginx (for serving the built React app)

Infrastructure

Container Orchestration: Docker Compose
Reverse Proxy: Nginx - Used to route requests to the appropriate service (client or server)
Environment Variables: Used for configuration and sensitive data management (e.g., OPENAI_API_KEY)

API Integration

OpenAI API (or similar LLM API) for advanced natural language processing tasks

Key AI Components

Text Embedding: Universal Sentence Encoder (USE) from TensorFlow Hub
- Converts text into 512-dimensional vectors
- Enables efficient semantic search and comparison of text segments
OpenAI Ada v2: A powerful text embedding model that enhances semantic search and question-answering accuracy.
Question Answering: Leveraging LLM capabilities through OpenAI API or similar services
- Generates context-aware answers based on the embedded text and user queries

This stack ensures high performance, scalability, and ease of deployment, making PDFGPT a robust solution for PDF-based question answering. The combination of USE for text embedding, OpenAI Ada v2, and advanced LLMs for question answering provides a powerful foundation for understanding and processing document content.

API Documentation

PDFGPT provides comprehensive API documentation using Swagger UI and ReDoc. After starting the application, you can access the API documentation at the following URLs:

Swagger UI: http://localhost:8000/docs
ReDoc: http://localhost:8000/redoc

How to Run

Prerequisites

Docker and Docker Compose installed on your system
Git (optional, for cloning the repository)
OpenAI API key
- You need to sign up for an OpenAI account and obtain an API key
- Visit OpenAI's website to create an account and generate an API key
Rename .env.sample to .env and update it with your OpenAI API key and other configuration settings.

Steps to Run

Clone the repository (if not already done):

$ git clone https://github.com/tuxxon/PDFGPT.git
$ cd PDFGPT

Build and start the Docker containers:

$ docker compose up -d --build

Note: The first time you start the containers, it may take 1-2 minutes for the backend to be fully ready. You can monitor the server logs to ensure it is running by using the following command:

$ docker logs pdfgpt-server-1 -f
 INFO:     Started server process [1]
 INFO:     Waiting for application startup.
 INFO:     Application startup complete.
 INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)

Once the containers are up and running, access PDFGPT:
- Frontend: Open a web browser and navigate to http://localhost or http://localhost:3000
- Backend API: Available at http://localhost:8000
To stop PDFGPT:
```
$ docker compose down
```