方寸 Portal

AI Engineer Portal

Your personal operating system for career transition.

Private mode
All Learning Paths
Learning Path

Evaluation and Observability

Measure behavior before scaling usage and complexity.

0 / 7 lessons complete0%
12h total intermediate
Continue where you left off

Evaluation metrics for LLM outputs

Choose the right metric for the right failure mode: accuracy, faithfulness, relevance, toxicity, and when LLM-as-judge beats deterministic scoring.

Start lesson