Language-AI Education Game
Raimi's AI Language Lab
A game-style web app where visitors explore how a language AI actually learns, by playing. With the museum character Raimi they work through five touch mini-games, each built on a real NLP concept: tokenization (slicing a sentence into pieces), embeddings (sorting words by meaning), word order, attention (spotting the word that matters), and next-word prediction with live probability bars.
The full flow (start, stamp-based lesson menu, play, completion) is plain JavaScript with no framework and no server; it supports both Korean and English (with browser auto-translate blocked so the exhibition copy stays intact) and ships as an offline-first PWA on the 3 exhibition tablets it runs on. Kiosk behavior is handled in-app: after 2 idle minutes it resets to the start screen for the next visitor.
A hidden, PIN-gated staff panel tracks per-device usage and exports reports to Excel through a dependency-free .xlsx writer, with a Node regression test covering the stats logic.
Highlights
- 5 NLP concepts as touch mini-games: tokenization, embeddings, word order, attention, and next-word prediction with live probability bars.
- Designed and built the full game flow: start, stamp-based lesson menu, play, completion.
- Runs entirely in the browser, plain JavaScript, no framework, no server.
- Bilingual Korean / English, with browser auto-translate blocked for exhibition reliability.
- Offline-first PWA on the 3 exhibition tablets, with a 2-minute idle reset for kiosk turnover.
- PIN-gated staff stats panel with a dependency-free Excel (.xlsx) export, covered by a Node regression test.
Architecture
How the app is structured: screens, roles, and data flow.
This site keeps anonymous visit statistics. Privacy