One of the core issue here is LLMs are probabilistic not deterministic, even with set of rules or context , llms are inherently built to make best guess, so my question is how you are ensuring that LLMs are following rules?