Huh, I just did basically the same thing. My requirements were not due to spending $300k/yr on parsing (lol), but I was amazed how far I got just asking the AI for progressively more functionality.

My use case is a bit different. I wanted JSONata as the query language to query Flatbuffers data (via schema introspection) in Rust, due to its terseness and expressiveness, which is a great combination for AI generated queries.