Perry - AI Powered Document Analysis. Perry allows you to upload PDF documents for question answering and information search. The project contains a fully functioning backend with authentication built in FastAPI, with document search built using LLamaIndex. The frontend has been built using Streamlit, and deployment is done through Docker.
You can upload documents, choose an agent to answer questions, and ask questions about the document. The agents are powered by OpenAI's API. Currently there is an echo agent for testing and a sub question agent that decomposes questions into sub questions for each applicable document. It is possible to implement more agents by adding them to the backend.perry.agents module.
Read the README is the subdirectories for more information on how to get started.
This repository can be developed and deployed using docker. The docker setup is split into development.yml and production.yml. The development file is used for local development and the production file is used for deployment to production.
A local docker setup:
docker compose -f docker/development.yml up
First go to docker/production.yml and enter your intended web host url and email for https encryption. Then run the following command to start the production docker setup:
A production docker setup with a reverse proxy and https encryption:
docker compose -f docker/production.yml up
This project has been created by Mickey Beurskens. Check my blog Mickey.Coffee ☕ or my company website at Forge Fire AI Engineering 🔥.
Everything is this repository has served as a proof of concept for a limited amount of users. You can use if for inspiration, but it is not production ready out of the box. Especially document uploads and session management will not scale with user demand.