LearningRobots

LearningRobots builds AI robots for teaching. The web app is how a classroom reaches them: it connects to the hardware, drives it, and turns whatever the robot is doing into something a child can follow.

Configuring a neural network in the browser

Students build the network in the browser. No install, no toolchain, no local Python. Whatever they put together is what the robot actually runs, which is the point. The model isn’t a simulation of the thing being taught, it is the thing being taught.

Two ways to reach the robot

Connection runs over Bluetooth or Wi-Fi, with SSL verification on the network path.

Classrooms aren’t uniform. Some have Wi-Fi a robot can join, some don’t, and a few have a network that technically works but blocks everything interesting. The app shouldn’t care which, so both transports land in the same place and it picks whichever is available.

Keeping the fleet current

Robot framework updates apply automatically from the web app on connection. Teachers never think about versions. A robot that’s been in a cupboard since last term catches up on its own the moment somebody plugs it in.

Showing the network learning

The teaching goal drove the visualisation work. I rendered the network so children can see how it’s wired and watch how it reacts, rather than hiding it behind a result and showing them the answer. The model is the lesson, so the model is what’s on screen.

Built to work without a network

The whole thing ships as a PWA with offline support. School connectivity is unreliable and a lesson can’t wait on it, so the app installs, caches, and runs with nothing behind it.