Sounds like just not using an LLM would be considerably less effort and fewer wasted resources.
It's a way to validate the LLM output in a test scenario.