CourseLarge Language Models: The Complete Course · Module 10: Evaluation · part 69 of 103
Part 69 · Module 10: Evaluation

Lesson 1: Build the eval before the feature

28 min read·13 Oct 2026
Module 10 · Lesson 1 · About 55 minutes · Needs: Module 4 Lesson 2 (the frozen split, run_prompt, score, load_split); Module 6 Lesson 1 helps (DraftReply)

The situation

It is Monday 3 August 2026, week 10. Shadow mode starts today: the assistant drafts every reply, and Lena or Ravi reviews each before it is sent.

Maya asks: "How will we know the drafts are good, other than reading a few?" Module 4 scored triage labels; a draft is open text. This lesson builds the eval for drafts before anyone tunes the drafter. The running example is T-1004: Yuki Sato bought the annual Team plan 20 days ago and wants her money back. The 14-day window has passed, so a good draft explains the rule and promises nothing.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.