A machine-learning teaching tool built for teachers, not students

I am building a tool for showing machine learning in a classroom.

Drag a point on the screen and a line follows it, so you can watch in real time how the computer works its way toward an answer.

I revised the plan three times, and twice it was not about features. It was about who the tool is for.

I re-cut the audience twice

The first audience was "people who find code hard." That is weak.

If you group people by interest, the reason to use the thing becomes "because I'm interested" — and interest alone does not keep anyone using a tool.

Cut sharply, the answer is this: a teacher who has to show something in class right now.

Teachers do not pick a tool by how many features it has. They pick by whether it removes hassle and risk.

Students are not the first audience. They are the ones who benefit.

What I am teaching has not changed. What changed is whose hands I put it in.

Decided and closed

• The first audience is teachers; students are the ones who benefit
• It must open in a classroom from a link, with no login and no install — this is the core constraint
• A big-screen demonstration mode is required
• Student management is pushed to a later stage
• Making money comes last — entering a competition and portfolio value come first

I built it two ways and threw one away

I built the same thing two ways and compared them.

• The web way — press a link and it opens. No install needed.
• The app way — good for putting on an app store, but using it in class means making people install it.

The moment you ask a classroom to install something, that lesson is gone. So I went with the web way. The app version was built only for comparison, then dropped.

Both versions do the maths without borrowing an outside machine-learning library. The point of the tool is to show what happens inside, so using a tool that hides the inside would contradict itself.

Still deciding

To stand out it needs depth beyond the basic picture. Three candidates:

• Training a model right inside the browser
• Looking at a learner's history and pointing out where they misunderstood
• Replacing the fixed explanation sentences with an AI that explains as it goes

What it taught me

The most regrettable blank is not a feature. It is the lesson script.

"When do you open this, what do you say, and what do the students do?" — that is still missing. And it is exactly the thing a teacher judges a tool by.