Live

Codify: build the eval suite

This is where engineers go deep and Builders hand off to you. Codify means turning your prioritized failures into automated checks: code-based assertions for things a rule can verify (format, presence, constraints), and LLM-as-judge for subjective qualities a rule cannot.

The critical discipline with LLM-as-judge is validating the judge against human labels - an unvalidated judge is just another vibe check, so you confirm on a sample that the judge agrees with you before trusting it. You build a real eval suite for your capstone here.