What is Inhouse

Inhouse is a user-friendly and intuitive document assistant that helps you improve your productivity. Inhouse integrates seamlessly with your Google Drive, allowing you to ask questions about your documents using natural language. It answers your questions effortlessly and provides sources from your documents, making knowledge retrieval a breeze.

Built with:

How does it work

1. Go to Inhouse
2. Create an account
3. Upload files or give a sharing link to one of your Google Drive folders by clicking on "Sync Drive"
4. Shoot your questions!

Architecture

Demo

inhouse.mp4

How to run it locally

1. Git clone the repo
2. Setting up the frontend
   • Go to the frontend directory
   • Make sure to use node 18 (use nvm if you have another version)
   • Run npm i
   • Create a .env file and set VITE_API_DOMAIN to the url of the flask server
   • Run npm run dev will start the server
3. Setting up the backend
   • Go to the backend directory
   • pip install -r requirements.txt
   • Set up a MongoDB cluster and replace the connection string on main.py
   • Make the passage-ranking.py file a Modal instance (more info here)
   • Creare a .env file and set the variables OPENAI_API, DB_PASSWORD and APP_DOMAIN (url of the vite server)
   • Run python3 main.py
4. Boom! You should be good to go but if you face a problem, feel free to make an issue :))

Credits

This project was greatly inspired from AlignmentSearch made by Fraser Lee, Henri Lemoine and Thomas Lemoine.

We've basically used their OpenAI API prompt and tuned it a bit for our use case!