xianchengz / pdf-ai

Full Stack Application to chat with ChatGPT with knowledge from uploaded PDFs

PDF AI

Integrated with Langchain to utilize OpenAI API or Llama to read PDF.

Demo

Architecture

The project include several components and they can be visualized as below.

Key Components

Backend

Two major components for the backend server are Flask and Langchain. Flask is used to build all REST endpoints and Langchain is used to interact with LLMs.

PostgreSQL

Postgres is used to store metadata including user information and PDF metadata that user uploaded.

AWS S3

All uploaded PDFs are stored in S3.

Pinecone

Pinecone as a vector database is used to store vector embedding of each PDF document, which is then used for retrieving relevant contents from PDF based on user's questions.

Local Setup

Prerequisites

Make sure you have the following items downloaded.

• Python3
• Docker

Setup

1. create a .env file with global variables from env-var-list.
2. Create a virtual environment.

python -m venv .venv

3. Activate virtual environment with

source .venv/bin/activate

4. Start up Postgres and Redis instance with inv run-local-infra.
5. Initialize database with inv init-db.
6. Start the application with inv dev.
7. Open a new terminal (make sure to activate the virtual env), start celery worker with inv devworker