IndexStorm / git-rec-back

Backend code for GitHub Recommendation Extension

GitHub Recommendations Backend

This is an official repository for the backend part of GitHub Recommender extension

Link to extension: Chrome and Firefox

Repository with extension code

indexStorm/git-rec-ext

Motivation

We find it hard to explore the world of GitHub repositories and are afraid of missing out some useful repos. That is why we decided to build this extension to help people to discover new horizons of hidden GitHub gems.

Once you've loaded the GitHub repo page, the extension adds a list of similar repositories based on their titles and descriptions. We do not collect any data and open source this product to everyone.

Explanation

Using SBERT embeddings and scrapped repo's descriptions we have constructed vector embeddings for over 100,000 repositories. Once the client sends the description of the current repo, we construct a vector from it and then search for the nearest embeddings in the vector space. You can read more about vector cosine similarity. Constructed embeddings and scrapped repo's descriptions can be shared upon request.

Tech/frameworks used

The backend was built on Python 3.9 using:

• Flask for server
• SBERT for building repo's embeddings
• FAISS for fast cosine similarity search

Made by indexStorm

Follow on Twitter @index_storm

Authors:

• Mike
• Evgeniy