What strategies do you employ to guarantee that an LLM output strictly conforms to a specific JSON schema in a production environment at Reducto?
