Study Buddy - Your AI Tutor

This project is in pre-release, we are looking for help on getting this into a first release ready state!

Our plans for this project is to make a fully offline and easily customizable RAG system to aid in research/studying. You control your data. Make use of your hardware.

Requirements

Cloud

Yes, this contradicts what we said earlier about a fully offline RAG system. But hear us out, Google is offering llama3.2vision-90b for free. You probably can't run such a model locally, so why not make use of it?

In /backend/secrets, you need to place a service-account-key.json with access to Vertex AI and set the GCP_SECRET_PATH in /backend/.env to the name of the file.

This is required for uploading .pdf. We only support .txt and .pdf currently.

Hardware

This was developed on a M1 Macbook Pro with 16gb RAM. Your mileage shouldn't vary much with different hardware as we kept cross-platform support in mind and running ollama is your responsibility.

Software

Running ollama is your responsibility, you can set the OLLAMA_HOST in the docker-compose.yml file. It defaults to 11434 as usual. We haven't created any executables yet, so you'll need to run the project as a Docker application.

Docker Usage

1. Clone this repo with git clone https://github.com/mithranm/study-buddy
2. Set the OLLAMA_HOST in docker-compose.yml to a valid Ollama instance (we really should have named it OLLAMA_URL, will be fixed in a future update). You have to do this for the backend service and celery_worker service.
3. Rename /backend/example.env to just .env
4. Run sh docker-build.sh to build for the first time.
5. Run docker-compose up with or without -d
6. Navigate to localhost:9091 in your browser and begin!
7. Press Ctrl+C to stop application
8. Run docker-compose down without our with -v to either keep your documents or delete them.

Development

Requirements

Software

• Ollama (serving on the port you configure, default is 11434)
• Redis (serving on port 6379)
• Poetry (Pyenv recommended)
• NodeJS
• Tesseract

Hardware/OS

Again, this was developed on an M1 Macbook Pro. Having a powerful computer is required. We aren't even sure Windows users can help develop this project. Linux should be fine.

Instructions

We use Git Flow for development. Switch to the development branch create a feature branch off of it to start developing.

git switch develop
git flow feature start feature-name
# After doing your changes, commit
git flow feature finish feature-name

Please check if we have any feature branches in the repository already, and refrain from naming your feature that.

Running While Developing

Backend

0. Have python 3.11.10 installed through pyenv
1. Navigate to /backend
2. Run poetry install
3. run sh run-integration.sh
   • If you have a Windows machine, we aren't even sure you can develop this project. Please contact us.
4. Congrats, the gunicorn server is running on port 9090.

Frontend

0. Have NodeJS v20 LTS installed
1. Navigate to /frontend
2. Run npm install
3. Run npm start
4. Congrats, the react app is running on port 9091.

Unit Testing

Backend

IMPORTANT: These don't work anymore and we are overwhelmed with features trying to fix everything. But we have verified the program runs as intended (for the most part)

1. Navigate to /backend
2. Make sure you have run poetry install before, run it if not.
3. Run poetry run pytest

Frontend

WIP - We are looking into selenium for testing the frontend (we also need to make a good frontend)

Future Updates

• We plan to fully support CUDA once we get our hands on some NVIDIA hardware. This should allow stronger embedding models to be ran using CUDAExecutionProvider
• We want to allow OpenAI generic apis to be configured as the model provider to avoid reliance on Ollama. Should allow usage of frameworks like vLLM and the OpenAI api itself.