what is really nice about this one is that:
a) don't overestimate LLM b) cache the right answer so you don't call the LLM api everytime c) fast enough
a) don't overestimate LLM b) cache the right answer so you don't call the LLM api everytime c) fast enough