Yeah, I do think that's possible with LLM, just too slow and expensive to be usable in most settings.