We did use an LLM for some of the composition of the ontology, then had a PT review, but we evaluate and deliver everything via deterministic logic.

We use some some LLM around the edges to process natural language but that's it.