Figuring out how to scale it may be a challenge, but it may be worth considering something along the lines of the directory sites that used to be the main method of discovery on the web prior to the rise of search engines, which itself can be searched.

I would personally find this useful at least, particularly when diving into unfamiliar topics. In a lot of those situations, a handful of high quality links can preempt a litany of more granular searches.

This might be an area where a traditional, respectful search engine style crawler backed by a small cheap LLM paired with human curation would work well. The crawler+LLM finds sites, attempts to compute some quality score, figures out where they'd fit in the directory's hierarchy, and then submits them for human review and approval. If a poor quality link makes its way through, users can flag it for review and if curation agrees it gets booted.

I'm leaning against the user flagging model of content moderation because it keeps the incentive for bad actors to try to game the system. Bad content, even briefly included can still do a lot of damage. It's like spam email -- even though something like 99.9% gets filtered out, that 0.1% is still valuable enough for spam senders to keep trying. If the problem of spam is just a numbers game, then bad actors will just increase the volume to compensate.