LLM-as-a-Judge: Evaluate Your Models Without Human Reviewers

Human eval does not scale. LLM-as-a-Judge matches human agreement rates at 1000x the throughput. Here are 3 patterns with working Python code.