Catch LLM regressions before your users do — a tiny CI gate for LLM output

You ship an LLM feature. Weeks later you tweak a prompt, or the provider rolls the model forward...