OpenAI Cookbook: Official Recipes · Lesson 6
Hallucination Guardrails for Support Agents
Build an output guardrail for customer support: generate an eval set with GPT-4o, create a judge model to check whether responses follow policies, and measure precision and recall.