Most teams that put a language model into a production path begin with the same pattern: write a prompt, ask politely for JSON, parse…