Pretty interesting. I like the prompt rewriting approach they have. I think I also saw they use the ontology to have the LLM generate examples (for context) that match the user set ontology, which is nice and makes sense.

→ View original post on X — @yoheinakajima