Embeddings based RSS recommendation system

About a year ago I created rssfilter to help me out catching up with my RSS feeds.

I was getting too many new posts per day from HN and other sites and started not enjoying it as much. I wanted to have 50% of the posts but missing out as little as possible about the topics I'm most interested in.

After a year using it, I'm pretty satisfied about the results and other users are too.

How rssfilter works is by being a proxy to your existing RSS subscriptions. It replaces the article links to a redirect that stores what articles you open. This information is what's used for the recommendations.

The embeddings of your read articles title and description are clustered. And from then on, you will only get 50% of the articles from all feeds, mostly what's closest to your areas of interest but also some percentage of unfiltered posts to allow for discovery.

It can be self hosted and there is also a live version that I host. No registry, the user ID is just a random UUID.

You can also use the API (FastAPI btw) directly: https://rssfilter.sgn.space/api/docs

Let me know if you like it!