Wouldn't something like snapshot testing from a scheduled probe be more effective and reliable than using an LLM?
Every X hours test the endpoints and validate the types and field names are consistent... If they change then trigger some kind of alerting mechanism to the user.
if the types and field names change, our parsing script should be able to detect that so it should be covered. I was talking about handling the subtle changes that are undetectable by checking field types and names
