Applied AI engineering program
Most AI courses stop at the demo. Keel starts with a real kind of problem: Lantern Home, a fictional online shop, gets about three thousand messages a month about parcels that never arrived, damaged kettles and refunds, and each customer waits two to three days for an answer. Across 13 phases you build the system that reads those messages, applies the shop's rulebook, settles the clear cases and sends the hard ones to a person, from your first line of Python to a deployed, audited service. Every submission runs against automated checks, and it passes only when the work holds up.
One payment opens Phase 0 and Phase 1. Nothing renews. Unit 0.1 is free to read before you pay.
13
Phases, one system
889
Hours of real build work
1
Payment for Phase 0 and 1, nothing renews
0.1
Unit free to read before you pay
How a unit works
Each unit is a small build with the same heartbeat: learn the concept, practice on a parallel task, ship the deliverable, get it graded, move on. No surprises after unit one, because the surprise budget goes to the ideas, where it belongs.
One written lesson, read start to finish. It opens with the concept, grounds it in the client's numbers, then shows exactly what you are about to produce, with checkpoints you answer before you peek. (Peeking is allowed. Peeking first is noticeable.)
Every chapter builds the program in front of you, a few explained lines at a time. Then you finish it yourself on the client's files, and automated checks tell you exactly which step is missing.
Upload the deliverable on the unit's page. The deliverable and the submission contract are published up front, so you will never have to guess what 'done' means.
Submit, and automated checks run your code. Each check says what it looked for, so a Not yet tells you what to fix.
Pass a unit and move to the next one. Only verified work counts, and your dashboard keeps the record.
Graded for real
Every result comes from automated checks you can read before you start. What happens when you upload:
Your program runs on our server, in an isolated container with the unit's own files, so the result does not depend on your computer. Each check reports Passed or Not yet and says what it looked for, so a Not yet tells you what to fix.
Where a unit's program reads customer messages, one check runs it on a message you have never seen. A program that prints answers it remembers, instead of finding them, does not pass.
No language model grades a unit today: the checks decide. Before any unit is graded by one, we will test that grading against work people have graded by hand.
Support on every unit
Stuck is normal. The lessons show you what goes wrong on purpose, and every unit has a tutor that has read its lesson.
Ask why something works, ask for another example, or paste the error that has been on your screen since midnight. It has read the whole lesson, explains basic Python the lesson leaves out and tells you so, and will not write the program you hand in.
Where a lesson teaches code, it breaks it on purpose: you see the traceback, learn to read it from the bottom up, and fix it before you meet it on your own.
Curriculum units
Unit 0.1 is free to read in full. Every other unit's page shows its outline and how many checks it runs before you pay.
none
Open the unitnone
Open the unitread_message.py, which opens messages/M-1041.txt and prints the customer's email address and the order number found in the message.
Open the unitCreate an account, take the 20-minute placement check or skip it, and open your first unit in the same session.