Building a vocabulary app that makes learning Korean and Thai feel more like a game than a grind

Vocabulary acquisition is one of the best-understood areas of language learning. Spaced repetition works. Active recall works. Multi-modal exposure — reading, hearing, producing — works. The problem isn't that we don't know how to learn vocabulary. The problem is that most apps implementing these methods are either too academic to sustain daily use, or too gamified to build real competence.

I was studying Korean and wanted an app that took multiple study modes seriously — one that would let me switch between flashcards when I wanted passive review, quizzes when I wanted active retrieval, and spelling drills when I wanted production practice — all from the same deck, without creating separate study sessions for each mode.

I also wanted to build vocabulary battles with friends studying the same language, because competitive learning is genuinely fun and fun is what makes practice sustainable. Nothing on the App Store did all of this offline, without a subscription, for Korean and Thai alongside English.

The core architectural decision was making decks the primary object, not study sessions. Every piece of functionality — flashcards, quizzes, pronunciation, spelling, multiplayer — operates on the same deck without any re-configuration. You browse your decks, choose a mode, and start. The session picks up where the adaptive algorithm thinks you need it most.

Adaptive repetition runs underneath every mode. The spaced repetition algorithm tracks performance per word per mode — so a word you recognise easily on flashcards but struggle with on spelling gets surfaced more in spelling practice, not across all modes equally. This is a more granular model than most vocabulary apps implement.

Each study mode targets a specific type of memory engagement. The design challenge was making four distinct interaction patterns feel like variations of the same system rather than separate features stitched together.

The progression system — XP, streaks, Beginner to Advanced levels — is designed to reward consistent practice rather than raw score. A user who does twenty minutes of spelling practice at 60% accuracy earns more than one who does five minutes of easy flashcards at 100%. The metric is time-on-task quality, not vanity correctness.

Multiplayer was the feature I was most uncertain about shipping — not because of the UX, but because of the infrastructure. Real-time synchronised sessions typically require a backend. I wanted to avoid a backend entirely, because a backend means recurring costs, maintenance burden, and — most importantly — a dependency that kills the feature if the service ever goes down or the project becomes dormant.

The solution was Apple's MultipeerConnectivity framework: peer-to-peer networking over local Wi-Fi or Bluetooth, with no internet connection and no server. Two players on the same network connect directly. The session host broadcasts questions; both devices render answers simultaneously; results are synced at end-of-round.