Learn by building things that work.

Every course ends with evidence you measured yourself, on your own task and your own hardware. Start anywhere — the free modules need no account.

15 courses · 116 lessons · 45 free, no account · 77 hours of work

Start here

Coffee break: the summariser you can prove is wrong — Twenty minutes to build it, twenty more to find out where it fails. The second half is the point. (2 free lessons)
Friday night: make one lamp answer to you — Voice in, relay click out, and the millisecond count that says where the wait actually lives. (3 free lessons)
The economics of automation — Why one job gets automated and the near-identical one next to it does not. (3 free lessons)

Course catalog

Atoms, not pixels — Get a model out of the browser and onto an arm that costs less than a phone. (29 lessons, 1615 min, 5 free)
Automate a local business — Somebody else's business, somebody else's process, and a handover where you never get called again. (23 lessons, 1010 min, 4 free)
Build an autonomous greenhouse — A closed loop that runs for six weeks unattended, in a system where failure is measured in dead plants. (3 lessons, 100 min, 1 free)
Coffee break: the summariser you can prove is wrong — Twenty minutes to build it, twenty more to find out where it fails. The second half is the point. (2 lessons, 40 min, 2 free)
Start a community robot lab — A room, some kit, and six sessions that other people finish. The hard part is not the robots. (3 lessons, 95 min, 1 free)
The economics of automation — Why one job gets automated and the near-identical one next to it does not. (3 lessons, 65 min, 3 free)
Where the energy actually goes — Joules, not vibes. Measure what your own work costs before joining an argument about data centres. (3 lessons, 65 min, 3 free)
Friday night: make one lamp answer to you — Voice in, relay click out, and the millisecond count that says where the wait actually lives. (3 lessons, 85 min, 3 free)
What humanoid robots are actually for — Count the cuts in the demo video, then count the contacts in your own hands. Two numbers, one honest picture. (3 lessons, 70 min, 3 free)
Local or nothing — Run open-weight models on hardware you own, and know what they cost you in watts and seconds. (5 lessons, 6 min, 2 free)
Price the work — Scope it, cost it, quote it — and know the number below which the job loses you money. (3 lessons, 80 min, 2 free)
Sell the thing, not the hours — Why your rate has a ceiling, where the repeatable part of your work is hiding, and what a product costs you after the sale. (3 lessons, 85 min, 1 free)
Ship one automation — Take one real process end to end, put it in front of people, and measure what it did for six weeks. (26 lessons, 1255 min, 8 free)
The eval you actually need — Stop reading leaderboards. Build the harness that measures your task, on your inputs. (4 lessons, 8 min, 4 free)
Validate it in 48 hours — Find out whether anybody wants it before you spend six weekends finding out they don't. (3 lessons, 65 min, 3 free)

Latest writing

How to measure LLM latency: find which leg owns the two seconds — A voice command that takes two seconds to fire has four legs and only one of them is slow. Here is how to time each one separately before you optimise anything. (2026-08-03, 4 min)
You don't need a benchmark, you need five probes — Public evals measure the average task. Yours has awkward inputs and one definition of wrong. Building the small, specific eval that catches your failures in an afternoon. (2026-07-28, 4 min)
What actually fits in 8GB of VRAM — The arithmetic behind model size, quantisation and context length — so you can work out what runs on the GPU you already own before downloading 40GB to find out. (2026-07-21, 4 min)

All writing · Browse by topic

Find your way around

Journeys — Pick the future you want to be part of, then follow the courses that get you there in a deliberate order.
The skill tree — Every concept the courses teach, drawn from what each lesson claims to need.
Evidence from other people — Measurements published by consent, with the hardware and the limitations attached.
Peer review — Put a result in front of somebody else against the same rubric they will be judged by.