This is THE ONE domain where you would want to use classical machine learning and not unreliable LLMs. Unless you want to kill yourself, that is.