If just recommending npubs suffices, then Plebstr kind of does this. Probably not even close to what you are looking for though.
If you are talking about an algo, then it's on us, meaning either that we have to provide that manually rn, or that we should rely on organic content recommendation or thst it's on us to develop that functionality.
Personally I am a fan of regularly diving into Global looking for new and interesting content and npubs to follow.
Global works sometimes for now, especially of you only use paid relays..
Ultimately though, you'd hope nostr will have a ton of content out there.. and sifting through it would be prohibitive.
I am thinking some kind of algo.. but like client side... am algo that enables the client to pull more of what they want more efficiently, not push where you see what somebody else wants to show.
How do you imagine that working? Meaning what would that algo base it's recommendations on? (or tbp how it would determine what is it thst you want to see?)
Umm.. honestly I'm not sure.. IRL I work with alot of different models, but mostly risk models.. I have little to no experience with language models.
However, content recommendation algorithms can be quite useful, Netflix is really good at recommending shows I might like. Pandora good at recommending songs I might like. Etc.
I think the danger is a single algorithm out of user control pushing the narrative top down... I imagine that if there were many competing algorithms that clients could kind of download and plug and play in to their client they could be a really useful tool.
