Full-stack engineer
building AI products.

HSKStory — Chinese learning

Built and run independently across web, iOS, and Android. Graded Chinese stories that editors tailor to each HSK level and check word by word against the HSK 3.0 lists, with narration, tap-to-define vocabulary, and spaced-repetition practice.

• AI narration costs. Where stories use AI narration, batch-generate it with open-source CosyVoice on rented GPUs and serve the finished files without live inference. Validate pacing, loudness, and audio quality before publishing.
• Web, iOS, and Android. Share React reader components between Next.js and Capacitor. Extend native audio for background playback and lock-screen controls.
• Offline learning. Support downloaded stories, audio, and flashcard reviews; reconcile progress, vocabulary, and study state when readers reconnect.
• Subscriptions across platforms. Integrate Paddle and RevenueCat for web and in-app purchases, with shared backend access checks across devices.
• Language quality. Validate HSK vocabulary, resolve Chinese word boundaries and pronunciation, and align narration to text for word-by-word highlighting.

CourtMix — Doubles scheduling

A React Native app for running doubles sessions courtside. Generate matchups, record scores, and track standings, with session data stored locally in SQLite.

• Refined through real play. Real-world testing exposed repeated partnerships. Increased their scoring penalty and added recency weighting to discourage back-to-back pairings.
• Fairness as players arrive. Adjusted late-joiner scheduling so an empty match history would not give new arrivals priority every round.
• Competing constraints. Balance team skill, partner variety, and rest across multiple courts, with mixer and Swiss-style formats.

cliMeter — AI coding usage at a glance

A native Swift macOS menu-bar app that keeps Claude Code and Codex usage limits visible while you work. User feedback led to Codex support and peak-hour information with a countdown in the user's local timezone.