I see you've given up. I haven't. LLM inside deterministic guardrails is a pretty good combo.