GitFeed
Discovery engine (in progress): a GitHub repository feed ranked by your interests, with search that understands what you mean.
Python · GitHub API · Embeddings · Semantic & Agentic Search · Recommender Systems
- Where
- Personal project
Problem
Finding a good repository on GitHub still depends on already knowing its name. Trending is the same list for everyone, and keyword search only works when the project's README happens to use the words you typed — so useful work in your niche stays invisible.
Solution
Model each user's interests from their own GitHub activity, then use that profile twice: to rank a personal feed of repositories, and to re-rank results from an embedding-based search index that matches on meaning rather than exact keywords.
Architecture
Overview
GitFeed is a discovery layer over GitHub. It builds an interest profile from signals you already leave behind — the repositories you star, the languages and topics you write in, the projects you contribute to — and uses it to rank a personal feed of repositories worth your attention, rather than a global trending list that looks the same for everyone. Alongside the feed is a search engine meant to be more efficient than keyword matching: repositories are indexed by embeddings of their README, topics, and metadata, so a query like "lightweight job queue for Postgres" can find a project that never uses those exact words. Results are then re-ranked against the same interest profile, so two people running the same search see results ordered for them.