Flagship programme Foundations of large language models
Eleven weeks on what a model can do, what it cannot, and how to put one inside an existing process without breaking what already works.
Understand the machines, then lead them.
Short taught programmes built with people who ship AI systems for a living. Eleven weeks, one real project.
Start for free 01 — The catalogue Four families,
Four families,
one standard.
Every programme is written by someone who does the job they teach. Nothing goes live until it has been run through a full cohort and rewritten afterwards.
Flagship programme 94%
of enrolled learners finish. The sector average sits near 13 %.
Applied data analysis
- SQL and modelling
- Applied statistics
- Reporting and decisions
“I dropped out of two courses before this one. Here the pace was survivable.”
Systems and deployment
- Serving and latency
- Evaluation harnesses
- Cost and observability
AI product design
Research, prototyping, user testing and handover to the engineers who build it.
All 87 programmes 0
Active learners
0
Programmes running
0 %
Completion rate
0
Practitioners teaching
02 — The method
How a cycle runs
Four stages across eleven weeks. The calendar is fixed in advance and does not move.
Scoping call
Thirty minutes with a lead instructor to place your actual level and rule out the programmes that would waste your time.
Taught weeks
Four hours a week, arranged how you like. One assignment every fortnight, marked by the person who wrote the module.
Project on real data
A brief from a partner company, or your own if you have one. No invented exercise at the end of the cycle.
Review and certificate
You defend the work in front of two practitioners. The certificate records the assessed brief, not attendance.
03 — Feedback What people say
What people say
six months on.
Collected half a year after the cycle ends, not on the way out of the last session.
“I had shipped models before but never had anyone read my evaluation code line by line. That single habit changed how our whole team ships now.”
Alice Duport
ML engineer
“The retrieval module was blunt about what does not work. I stopped three projects at my company on the strength of it.”
Marc Leroy
Data lead
“A working community and a brief with real constraints. That was what I needed after a long career break.”
Sophie Martin
Product designer
“The scoping call moved me to a different programme before I had paid anything. Few schools talk you out of their own product.”
Thomas Vasseur
Engineering manager
04 — Pricing
Two plans, cancel anytime
Cancel from your account in one click. Programmes you have started stay open for thirty days afterwards.
Explorer
$0 per month
- Nine introductory programmes
- Cohort forum access
- Email support within 72 h
Full cycle
Most chosen $19 per month
- All 87 programmes
- Every assignment marked by hand
- Project on real data and review
- Certificate recording the assessed brief
- One mentor call each month
05 — Questions
What people ask most often
For as long as your subscription is active, with no time limit, updates included. After you cancel, any programme you had started stays open for thirty days so you can finish a module in progress.
It records an assessed brief reviewed by two practitioners. It is not a state-issued degree, and we say so plainly on the scoping call rather than leaving you to find out later.
Card and direct debit. For employer funding we issue an agreement and an invoice in the company name, and the finance team never needs to speak to you about it.
Yes, within thirty days of the first payment, with no justification required. Refunds are processed within five working days to the original payment method.
Not for the introductory programmes. For the advanced cycles the scoping call exists precisely to check prerequisites and, where needed, point you at an intermediate programme first.
06 — Journal
All articles
Notes from the teaching floor
Write the evaluation harness before the model
Teams that build the test suite first ship slower for two weeks and faster for the following year. Here is the order we teach.
A long context window is not memory
Larger windows changed what fits in a prompt. They did not give the system a memory, and conflating the two produces a specific class of bug.
Write the model documentation before you need it
Every team writes documentation after their first incident. The work is the same either way; only the circumstances get worse.
January 2027 cohort Ready to start
Ready to start
your cycle?
Enrolment opens on 6 November. Thirty places per cohort.
Start for free