What's the best way to explore a large set, one by one? Maybe cluster names (maybe there's a meaningful-enough language model that works), then when someone likes a name, sample within cluster n times, before hopping out to the big set again.