Choose a sorting problem, build a fair test set first, train and improve a Teachable Machine model until every group passes, connect a micro:bit build that acts safely on its labels, and report what it still gets wrong.
Engineers who build an AI system are asked two questions before it is used: how did you test it, and when does it fail? This time the problem, the data and the answers are yours.
You have trained a model, tested it fairly, fixed it by changing its data, and sent its labels to a micro:bit. Now do it all on a problem you choose, in four milestones with less help each time.
Some ideas: ripe or unripe fruit; LEGO bricks by colour or size; tidy or messy desk; book or not a book; wet or dry washing (photos only); a recycling sorter for your school's own bins. Two to four classes, plus a Neither class for everything else.
Make your prediction first. Help opens once you've locked it in.
Write in your book or a document:
Take your test photos before you train, and keep them apart from your training photos. For each class, take photos in at least three groups: different places, light or distances where the model will be used. On the board, choose the class, name the group, then Add photos. (The photos stay in this browser only.)
Add at least 24 test photos in at least 3 groups, with every class in every group.
Use the board. This ticks when you get there.
Train in Teachable Machine (new photos, not your test photos), export with Upload my model, and paste the link. Find the weakest group, work out what its failures share, change the training data, and test again. Keep a note of each change and its scores.
Get 80% or more overall, with no group below 60%.
Use the board. This ticks when you get there.
Program the micro:bit: one branch for each of your labels, an else that does something safe, and a reply (serial write value) saying what it did. A servo flap, lights, a sound or the screen: your choice. If the build is across the room, add the radio bridge.
The micro:bit acts on each of your labels, does something safe for anything else (Unsure, Neither, or a word it doesn't know), and replies with what it did.
This ticks by itself when your blocks do what is asked.
Load your model, connect your micro:bit, and run it on the webcam with real items. Watch what happens when you show something that is in no class at all.
Run the whole system: at least 5 labels sent from the webcam and 5 replies from the micro:bit.
Use the board. This ticks when you get there.
Write a one-page report (or a two-minute talk):
Students build an AI sorter for a problem they choose, in four milestones (test set, model, build, system), and report its scores by group and its limits. About 120 minutes.
Open the teacher guide: answers, solutions, slides and a worksheet.
The capstone brings the course together on a problem each student (or pair) chooses. The order matters: test set first, from photos the model never trains on, in groups that match where it will be used (place, light, distance). The board keeps the test photos in the student's browser only (testset: own); they are shrunk to small squares and never uploaded. Remind students not to photograph people.
Milestones: (1) at least 24 test photos in at least 3 groups; (2) 80% overall with no group under 60% (the board's acc_min); (3) a micro:bit program with a branch per label, a safe else and a reply; (4) the whole system on the webcam (typical: needs a micro:bit). The report is the assessment piece: scores by group before and after, the data changes made, and the limits.
Choosing problems: steer away from people (faces, emotions, who someone is) and anything where a wrong answer could hurt someone. Good problems have objects students can bring in. Two to four classes plus Neither is plenty.
Without micro:bits: milestones 1, 2 and 3 work without hardware (3 is checked in the editor); milestone 4 is typical.
Making the test set before training and keeping it separate. When the score is 100% at once, ask whether any test photos were also training photos.
Stretch: Add a second model in CreateAI (a movement) that starts or stops the sorter, so the two models work together.
Support: Offer a ready-made problem (the recycling sorter with your school's bins) and a planning template with the four plan questions.
Where this leads: AI robots (a robot driven by a model, with safety rules), and the Python pathway's Machine learning with Gradio. Real world: model cards and test reports for AI systems.