Building teaching tools with GenAI
The ink is invisible. The exercise is yours.
A university teacher can now build an interactive exercise that paper can't do, in an evening instead of a week. GenAI is the partner at the workbench: it suggests the lines, and the teacher draws them.
The students never meet it. What reaches the classroom is an ordinary web page, with no AI, no accounts and nothing leaving the browser.
00The exercise
01Why
02Workflow
03Tech
04Cost
05Privacy
06Quality
07Catalogue
08Get started
—What students see
The exercise
In the middle is what is being built: a sketch of an exercise. The stations lie around it, each with its own tool.
The workbench photo is AI-generated (GPT Image 2.5 via Higgsfield). The exercises it points to are not.
Eight stations
Everything it takes, one question at a time
- 01WhyWhat can an exercise on screen do that paper cannot?
- 02WorkflowFrom a teaching idea to an exercise in use.
- 03TechWhat do I need, and how much do I have to know?
- 04CostSubscription, hosting, domain, and the real cost: time.
- 05PrivacyWhat leaves the student's browser, and what the GDPR asks.
- 06QualityAccessibility, subject accuracy and trying it on students.
- 07CatalogueExercises in use, with small examples to try.
- 08Get startedA template to take home: README, CLAUDE.md, deploy script.
Station 01 · The paper test
If paper can do it, use paper
An exercise earns its screen only if it uses at least one thing paper can't do. There are seven, and each keeps the name the research already gives it.
- POEPredict–Observe–ExplainYou commit to an answer before you may see it.
- A/BSplit-class experimentThe class is split into conditions without knowing.
- RTReaction timeTime is measured in milliseconds, exposure is controlled.
- ARSAudience responseThe class's answers are collected and shown live.
- SRSSpaced retrievalThe page remembers you and comes back weeks later.
- SIMInteractive simulationTurn a dial on a model and see what follows.
- RWDReal-world dataWork in real material that can't be printed: CT stacks, X-rays, signals.
Stroop = RT + ARSAnchoring = A/B + ARSSegmentation = RWD + SIMMCQ = ∅MCQ + ARS = Peer Instruction
Build time
Built with GenAI, and it shows in the history
Every exercise in the catalogue lives in a git repository. The numbers below are counted from those histories each time this site is published.
“An exercise that used to take a week of programming now takes an evening.”
31 seconds
From the bench to the class
At the bench, hands of light help and the ink suggests the lines. The pencil traces them, and the glow goes out. In class, nothing glows.
AI-generated film (GPT Image 2.5 and Kling 3.0 via Higgsfield). No sound.
In use
Four sites, four audiences
- ex.samle.dkEU medical device regulation, risk management and signal processing for health technology students.University · in English and Danish
- insights.samle.dkForming teams and running group work, as tools in the browser.University · workshops
- psykologi.samle.dkTwenty psychology exercises: reaction times, hidden splits, class results.Upper-secondary · in Danish
- 3dklinikken.samle.dkMedical 3D printing for 13-year-olds, from real CT slices to a 3D model.7th grade · in Danish