Build Your First AI Eval

Every session is recorded, and the recordings and resources stay yours for life. That said, the course is built around tagging, grading and running in parallel with me, then getting feedback on your work. Watching the replay alone is not the same experience, so plan to attend live when you can.

Most evals courses are deep dives: four weeks, several hours a week, a price above $4,000. This one gets you to your first eval in three sessions, on one live example, under $500. No vendor pitch: I teach with Claude Code, but the method works with any tool. Think of it as the on-ramp: the deep dives make more sense once you have run one eval.

No, and you can follow every session having never opened it. I teach with Claude Code because it makes the concepts easiest to apply, and we set it up together in session 1, from zero. What you learn is the method (golden set, criteria, judge, score), and it works the same with Codex, Cursor or any eval tool your team already uses.

None at all. Sessions 1 and 2 have no code: you write criteria, tag data and read scores. Session 3 is vibe coding: you describe the tool you want, Claude Code writes the code, you direct and test it. If you can write a Slack message, you can drive it. I give written, personalized feedback on your assignments so nobody gets left behind.

If you want ML theory, or evaluation strategies for RAG, multi-agent or multimodal systems, this is not that course. It is your first eval, end to end, on one simple use case. It is also not for people who expect a developer to build their evals for them: you build yours, and you leave able to explain it.

One 90-min live session per week, plus 1 to 2 hours for the assignment on your own AI feature, plus about 30 minutes if you want to rewatch a replay. Three to four hours a week over two weeks. The assignment is optional, but it is where the method sticks.

Excalidraw is an open-source whiteboard app for hand-drawn diagrams, with 130k+ stars on GitHub. Its issues are public, real and messy: bugs, feature requests, support questions. That makes it a perfect training ground: everyone works on the same data, nothing is confidential, and the method transfers to any product.

Most students expense it through their learning and development budget: evals are a team skill, and you bring back a method, a boilerplate and a tool. Click "Get reimbursed" on this page to generate a pre-filled letter for your manager. You get a receipt on enrollment.

It is enough to build a real eval and short enough to actually finish. Session 1 is your first eval end to end, session 2 adds a judge for free-text output, session 3 gives you a tool to keep running it. The complex stuff comes after, on your own features, with the momentum this gives you.

Yes. You receive a certificate of completion after the final session, to share with your employer or on LinkedIn.

Yes. I run private team trainings, offer group discounts when 3 or more people from the same company enroll, and make bundle offers with my other courses. Email me at jules@vibecodingacademy.ai and we will set it up.