This is neat! What I would do is I would have a list of tags and make sure every post gets pigeonholed into those because otherwise it might be hard to match tags if extracted on a per-post basis (e.g. I see `large-scale data systems` for a job listing, which will rarely match a candidate because it's so wordy).

Cool product though.

This is something I'm considering. I was thinking we could probably hum it in with plain old vector embeddings, maybe?