Little Explorers

The setup was a nightly cron job: a headless AI coding agent given a topic brief (how the water cycle works, what insects do, why we have seasons) and a shared design system — big touch targets, read-aloud buttons, no text walls. Each morning there was a new app on the family hub, playable on the living-room tablet. Some nights it polished an older app instead of starting fresh; the cadence was roughly one new topic per night for a month.

Every app follows the same shape: a small set of hands-on stations (tap, drag, watch what happens), a read-aloud storybook, and pre-generated audio narration so nothing depends on a live speech service. The agent ran locally — local models and a local text-to-speech voice — and checked its own output before publishing.

These are AI-built teaching aids, not curated lesson plans. The content is accurate enough for a five-year-old's questions, and it sparked plenty of them, but it is machine-generated and unedited. The nightly builder job is now retired; what you see here is a frozen snapshot of the final builds, converted from their original local-server form to pure static files.

The apps

Ten topics, final builds. Tap a card to open the app.