Live: https://game-generator-three.vercel.app · Code: 02_ai-engineering-code/game-generator/ (scripts/ingest-flashcards.mjs, README.md)
Problem. Turning exam flashcards into varied, playable learning games by hand doesn't scale.
What I built. A generator that ingests flashcard specs (PNG + JSON metadata) and auto-produces game variants across 5 modes (match-answer, order-steps, what's-missing, multiple-choice, categorize) from 170 templates, with per-mode suitability scoring. A human-in-the-loop admin authors/curates; students play via an IndexedDB-persisted dashboard; playtest feedback is captured to playtest-feedback.jsonl / PLAYTEST-NOTES.md to improve future generation. Next.js 16 + React 19 + TS, Playwright tested at phone viewport.
Why it matters. A deterministic content factory with a structured feedback loop — the same eval/iterate discipline behind good AI systems, applied to learning-game generation. Genuinely novel product surface.
This one's unique. It takes flashcard specs and automatically generates playable games across five different modes from a library of templates, and it scores each variant for how well it suits the content. A human curates the best ones, students play them, and crucially the playtest feedback is captured to a structured log that feeds back into generation. So it's a content factory with an eval-and-iterate loop, the same discipline you'd want around an AI system, pointed at learning games. It's deterministic by design, which keeps it cheap and predictable.
GG · Applied AI Engineering portfolio · synced to youdontneedmy.help design