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.
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.