AWS docs have (had) an embedded AI model that would do this perfectly. I suppose it had better training data, and the actual spec as a RAG.