Comment by sburud

That’s cool! Slight fear of replicating the Dropbox comment here, but all you really need to do is run whisper (or some other speech2text), then once the user stops talking jam the transcript through a LLM to force it into JSON or some other sensible structure.