sheen robotics · teacher training handbookask sheen — the AI coding tutor inside the sheenbot canvas

Open the canvas (a logged-in project or the free, no-account canvas at sheen.bot/canvas-try) and you'll find a round blue button pinned to the top-right corner — that's ask sheen. Nothing to install; it works on the laptops, tablets and phones your classroom already has.

Clicking the button opens a two-step entry: step 1 picks who is coding, step 2 picks what to do. The age choice is remembered on the device, so returning students land straight on step 2.

The four modes at a glance

The student states an idea (say, "tell us when the classroom gets too hot") and sheen advances through six stages — Idea → Sense → Do → When → Plan → Build — asking exactly one guiding question per stage. Which sensor, what threshold, what happens when: every decision belongs to the student.

The menus are deliberate. Teachers told us that a blank box gets vague briefs — students grasp requirements far less readily than adults do — so sheen decomposes the project into concrete choices and lets the student pick. It still adapts to what they say:

• A specific answer moves on. No repeated questions for a student who already knows;
• A vague answer ("you decide", "dunno") earns exactly one finer question at the same stage, splitting the vague thing into three to six concrete readings — then the flow moves on regardless. sheen narrows the choice; it never takes it over;
• No idea at all? The very first question offers a starter-project menu instead of an empty page;
• Some questions accept several answers ("Pick one or more, then confirm") — for example two outputs at once — with a confirm button so a stray tap isn't a submission.

After the six stages sheen recaps the whole plan in plain language and asks the student to confirm — the student owns the design before a single block is placed. Confirming also picks the delivery:

Your starting scaffold

This is a starting point — not the finished program. Your job is to fill the gaps below.

initialize sheenbot∞

forever

set temperature to read temperature

if temperature > ? then

set onboard LED ?

+− drag to pan · 12 blocks

1

Your turn — fill these in

Fill in the "too hot" threshold

Think: at what temperature does your classroom feel too hot? Try a number between 28 and 32, then test it for real.

2

Pick the alert colour

Which colour says "warning" best? Why?

Before you build — explain it

• Why does the "if" check live inside the "forever" loop?
• What would the program do if > became <?

3

Add scaffold to a new project4

• 1The whole program is shown before it is applied — a real, read-only canvas you can drag to pan and zoom with + / −. What you see here is exactly what the button will place. (Very large merges fall back to an indented text outline of the same program);
• 2Every gap also comes with a real block picture and a hint that says what to think about, not what to type; tapping a block picture locates it on the canvas or in the toolbox;
• 3The "explain it" prompts are ready-made for pair-explaining or cold-calling — "you understand it when you can explain it". They appear on the complete program too;
• 4The scaffold always opens as a new project — it never overwrites the student's work. Before delivery sheen also test-runs the program in the simulator and silently rebuilds it once if it fails.

The preview matters most in "complete" mode, which by definition has no gaps to look at: the student reads the finished program before it lands on a canvas, which is exactly the moment to ask them what each block is for. Use it as the checkpoint — "before you tap Add, tell me what the third block does."

Classroom procedure

1. Have students describe the idea in one full sentence (board template: "I want to build a ____ that ____ when ____");
2. Students answer stage by stage — encourage reasons, not "right answers";
3. Pause at the Plan stage: each student reads their plan to a partner before tapping "Challenge";
4. Scaffold lands in a new project → fill the gaps → run on the board or simulator → answer the "explain it" prompts.

When a program doesn't behave, students pick this mode. sheen first asks them to describe in their own words what the program should do (often half the diagnosis right there), then helps in three escalating layers — one layer at a time, with the answer always last:

The conversation doesn't end at hint 3

Two things happen after a fix is opened, and both exist because a fix that doesn't work is the moment a student most needs the tutor:

• The typed box stays. Running out of hint levels only greys out the escalation buttons — the "tell me exactly where" box remains for the whole session. sheen even asks first: "I opened that as a new project — give it a run. Did it sort it out, or does something still go wrong?" Students used to reach for "Start over" here, which throws away the entire diagnosis;
• sheen may correct itself. Hints normally withhold the fix so the student finds the change themselves — but once a fix has been handed over and it was wrong, that reason is spent. From then on sheen can send a corrected fix instead of more prose about a program it just broke;
• The panel steps aside, it doesn't close. Applying a fix minimizes ask sheen so the student can see their canvas, and the pulsing pill holds the "did it work?" round — one tap comes straight back to it.

One tap turns the workspace into a step-by-step walkthrough: an overview first, then each step lights up exactly the blocks it covers — read aloud automatically for younger students. The walkthrough card floats at the bottom of the canvas, clear of the blocks. The last page is a challenge: a few questions about the program they just read (see The challenge below).

sheen.bot — coding canvas

1

Explain Step 3 / 5

Check whether it's too hot

This step compares the variable "temperature" with 30. Only when the temperature is above 30 does the block inside run — the onboard LED turns red. This is where your program makes a decision.

show on canvas

Prev Next

2

• 1The blocks covered by the current step glow on the canvas (here, the "if" check and everything inside it); the spotlight follows as you change steps;
• 2Card buttons, left to right: read aloud, re-fetch the explanation (use after editing the program), minimize, close; Prev / Next move freely in both directions.

The challenge: the page that asks them

Reading is passive, and a walkthrough that ends on "Done" never asks the student to produce anything. So the last page is a short challenge — two or three multiple-choice questions written from their own program: their threshold, their branches, their variable names. A question that would make just as much sense on someone else's project does not belong there.

The questions follow the age band: younger students get two questions with two or three short options, read aloud together with the options; older students get up to three, with one that tests order or logic rather than a single value. Wrong options are never jokes — they are the mistakes a student who half-read their code would actually make (an off-by-one threshold, the two branches swapped), which is exactly what makes the re-read worthwhile. And the canvas spotlight clears on this page: the questions are about the whole program, and a glowing block would point at the answer.

Five high-value uses

• Inheriting a project — when teams rotate or continue last term's build, a new member reads the code in minutes, not lessons;
• Learning from examples — open a class sample, hear it explained block by block, then remix it;
• Classroom review — project a student's build and let sheen narrate it to the whole class, blocks glowing in sync — far clearer than describing code out loud;
• Checking understanding — the challenge is a ready-made exit ticket: it tells you (and them) whether they understood the program or merely finished it. Ask for the score, or have them show you the question they got wrong first;
• Presentation rehearsal — before a demo or assessment, students hear their program explained once, then close the card and retell it themselves.

Students rarely stop at version one. The fourth mode takes any existing project — AI-scaffolded or built entirely by hand — and adds the next feature to it, without starting over.

There are two ways in: the “Add to my project” card in the mode chooser, or the “Want to add a feature next?” idea chips sheen offers right after a build — tapping a chip opens the next feature session with the goal pre-filled, so projects can grow feature by feature, lesson after lesson.

Add a feature togetherI'll help you think it through — you do the building.

GoalPlanBuild

What would you like your project to do next?1

Beep when it gets too hot!

Want to add a feature next?

Show the temperature on the OLED screen Green light when it cools down

Or type your own idea…

4

2

Added to your project — the new blocks are glowing on the canvas.

Undo this change

3

• 1The flow is Goal → Plan → Build: one question — what should the project do next — then a plan to confirm, exactly like a guided build;
• 2The feature is merged in place under a preservation contract: the student’s own blocks, values and wiring stay exactly as built, fill-in gaps land only on new blocks, and the new blocks glow green so the whole class can see what changed;
• 3“Undo this change” reverts the whole addition in one tap. If the blocks were edited after the merge, sheen asks first (“undoing removes those changes too”); in rare cases an amber notice invites a quick check of the blocks;
• 4After applying, sheen proposes two or three next ideas — and a box under them for an idea of the student's own, so the suggestions never box them in. Either way the next session opens with the goal already filled in: the chain that turns a one-lesson build into a term project.

What "add a feature" will not do

Merging into someone's own work is the riskiest thing sheen does, so three guards sit in front of it. Teachers meet them as short messages rather than as silent behaviour:

• It refuses to pretend. If a turn comes back without actually changing anything, sheen does not apply it and does not congratulate itself — it says "I couldn't add anything to your program that time — nothing on your canvas changed. Try again, or say the feature a different way." A student sent hunting for blocks that were never added cannot tell our bug from their own mistake;
• It won't quietly undo manual edits. sheen works from the canvas as it was when the conversation started. If the student edits blocks mid-conversation, an amber note appears before Add: adding now would restore the older version, and "Start over" is the way to work from what is on the canvas now;
• It won't write into the wrong project. Switch tabs mid-session and Add asks you to switch back first — a full-program merge landing on a different project would overwrite it.

Voice: read-aloud, speak-to-answer, hands-free

• Read aloud — the speaker button next to every question reads it out; the younger band auto-plays it, so reading level is never a barrier;
• Speak your answer — tap the mic and talk; a live transcript fills the input box. Great for slow typists;
• Conversation mode — fully hands-free: sheen reads the question, auto-opens the mic ("Speaking…" / "Listening…"), and a pause submits the answer.

Session management

Working language

• ask sheen speaks every language the canvas speaks — 11 of them, including isiZulu, isiXhosa and Afrikaans. It answers in whatever language the canvas interface is set to; switching the interface language switches the tutor with it;
• Read-aloud follows the same setting, with a warm neural voice per language (a couple of languages fall back to the browser's built-in voice — typing and tapping are unaffected);
• In Chinese the board is named as 赢星派Pro / 赢星派Mini, matching the hardware in front of the student; every other language uses the sheenbot names.

When the Wi-Fi drops

Classroom networks flap, and that used to destroy every open session. It no longer does: while the connection is down the launcher hides, any open panel stays exactly where it was with its inputs disabled behind an amber notice — "You're offline — reconnect to keep going. Your conversation is saved." — and it picks up mid-conversation when the network returns. Nothing to reopen, nothing retyped.

ask sheen does need the network, though: it is the one part of the canvas that cannot work offline. Block editing, running on the board and the simulator all keep going without it.

Block pictures and hardware awareness

• When sheen mentions a block it shows the real block picture (e.g. ) — tap it and the canvas locates the block, or guides the student to it in the toolbox;
• In debugging, that picture is the student's own block with their own values — "wait 3 seconds", "play Ring 2" — not a generic example. They are looking at the thing on their canvas, which is what makes "this block" unambiguous;
• Variable names draw as blocks too: when sheen writes about temperature, it shows the variable block rather than a grey code chip — so a student who reads pictures better than words can still follow;
• sheen knows the sheenbot∞ board: which modules are built in, which pins they occupy, which combinations conflict. A wiring plan that collides with an on-board module or the I²C bus gets flagged before the student hits run.

For students who have outgrown blocks

There is a fifth AI surface, and it plays by different rules on purpose. Open the Code Preview panel (the Python view beside the canvas) and ask sheen there, and it writes a complete, runnable MicroPython program — no staged questions, no gaps. It is a separate tool with its own instructions, kept apart from the tutor so the two contracts can't leak into each other.

Treat it as the graduation path, not a shortcut: a student who can read and edit the Python is past the point where a scaffold-with-gaps teaches them anything. A student who is not there yet will simply have a program they cannot explain — which the four block modes are specifically designed to prevent.

Feedback: help the AI get better

An AI that always says "done!" teaches students to stop checking. ask sheen is built the other way round: when a build is incomplete, when it may have touched more than it meant to, or when it achieved nothing at all, it says so at the point of decision — in the student's language, in plain words, right above the button. These notes are not error messages to tap past; they are the moment the lesson gets real.

The one-liner for parents: ask sheen doesn't write code for your child — it's an always-available coding coach whose first question is "what do you think?" Nothing to install; it runs on the devices the school already has.