AI engineering,
drawn — not described.

Long-form breakdowns of how real AI systems get built and measured. Every episode ships with a set of notes carrying the same diagrams that are in the video — so you can read what you just watched.

Read the notes Watch on YouTube

Latest

EPISODE 6 · RAG EVALUATION

The Eval Suite

You built a RAG chatbot for an airline. How do you know it works? Recall, precision, faithfulness, answer relevance, correctness, safety, speed and cost — what each one asks, where in the stack you measure it, and how the whole suite fits together.

Read the notes

1Componentsone at a time, on its own

2Pipelinethe RAG triad, end-to-end

3Applicationjudge the whole product

↻Regressionchange one thing → re-run the whole suite

4Onlinelive traffic · no answer key

↻Self-improving loopfailures become new tests

What you get

The episode

One idea per beat, animated. No stock footage, no talking head — every claim is drawn while it is said.

The same diagrams

The notes redraw the figures from the video, in the same colours. Cyan is always the retriever. Purple is always the generator.