10 · Build/think project: teach a smart device

Time to put it together. This is a build-and-think project — you'll design a small AI-powered device, and you can actually try the AI part for free, with no purchase, using an online tool and (optionally) a board you may already have at school.

No micro:bit? No problem. micro:bit CreateAI is a free, web-based tool with no login required (Micro:bit Educational Foundation, n.d.-a). You can plan the whole project and even collect movement data; running it on a physical board is the optional bonus.

Your mission: design a device that uses sense → think-with-AI → act, then plan it like an engineer.

Step 1 — Pick a job. Choose something that recognizes a pattern. Examples: a posture buzzer that recognizes "slouching" vs. "sitting up," a "good-throw vs. bad-throw" coach, a sign that lights up when it recognizes a specific gesture.

Step 2 — Name your loop. Write down your robot's three parts:

Sense — which sensor? (motion/accelerometer is easiest with CreateAI)
Think-with-AI — what two or three movements will the model learn to tell apart?
Act — what's the output? (a light, a sound, a message)

Step 3 — Plan your training data. List the labeled examples you'd collect for each movement, and — thinking back to bias — how you'd make them varied (fast and slow, different people) so the model isn't fooled.

Step 4 — Add a safety rule. Using Lesson 8, decide: what can your device do on its own, and is there anything it should flag or hand to a human? Write at least one hard rule for any higher-stakes action.

Step 5 — (Optional) Build it. With a micro:bit and CreateAI, "collect your movement data from the micro:bit's accelerometer, train, test and improve your own machine learning model, then use it" in a MakeCode program (Micro:bit Educational Foundation, n.d.-a). On an Arduino, the same kind of on-device machine learning is possible via TensorFlow Lite for microcontrollers (Arduino, n.d.). Either way, iterate: test, find where it's confused, add better examples, repeat.

Your engineer's checklist:

✅ My loop names a real sense, a learned "think," and an act.
✅ My training-data plan is varied, so the model isn't biased by one-sided examples.
✅ I wrote at least one hard safety rule and said when a human stays in charge.
✅ I planned to iterate — fix mistakes with better data, not by giving up.

Think about it. After planning (or building), name one thing the AI did well, and one thing you — the engineer — had to decide that the AI could not.

Sources

Micro:bit Educational Foundation. (n.d.-a). Artificial intelligence (AI) — micro:bit CreateAI (free, no login). https://microbit.org/get-started/features/ai/ Arduino. (n.d.). About Arduino (on-device ML with TensorFlow Lite). https://www.arduino.cc/en/about