custom recommendation engines that would work over the entire internet. similar to what parallel web systems is doing with search, but interest-based / user defined embeddings
I think if we combine nostr with mainstream content it will show the world that if every website used a native login, we'd be able to interact across all websites. so doing a ton of crawling/scraping of the entire internet essentially.
additionally reuse use the clustering process to create topical relays that would act similar to subreddits. classify each note + continuously run topic modeling on all notes to route notes to topical relays and clients could build dedicated clients on top