LLM evaluation metrics include RAG metrics like faithfulness and answer relevancy, agent metrics, and LLM-as-a-judge, explained with working DeepEval code examples.