I was working on something I thought would help but unfortunately had to focus on other things. Essentially wanted to use a combo of clustering algorithms + LLMs to topic model all of nostr and create topical relays and clients focused on particular types of conversations.

very similar to how subreddits work, but without any moderators. The clustering algos/LLMs would organize content based on embeddings (replacing moderators) and it would be much easier to find the stuff you care about without knowing who to follow.

clustering might be better bc it's not as "black box", but it's a great way to use machine learning without going down the recommendation algorithm slope