merk
A language course and flashcard app for adults, installable on your phone as a web app, with
invite-only accounts. Two courses, each with its own lessons, words and progress: German
A1 → B2 (for life in Germany) and Moroccan Darija A1 (Casablanca). The app itself is in
English or French; German is explained in English, Darija in French.
German
- Course: 125 lessons in parts A1.1 → B2.2: grammar, topic lessons (a theme's words) and
stories. Each lesson has four parts: the lesson itself, listen & read, speak, write.
- Exams: a short exam closes each part (70% to pass); a Goethe-style exam closes each level
(listening, reading, grammar, writing, speaking; 60% to pass). The path waits at an exam.
- Placement test: words, grammar and listening, about 10 minutes; starts you at the right
lesson.
- Flashcards: ~4,600 words and phrases, A1 → B2, scheduled with FSRS
(ts-fsrs). der/die/das colours, typed
answers that accept
ae/oe/ue/ss.
- AI teacher: feedback on your writing, and on spoken answers in level exams (Claude Haiku
via Vercel AI Gateway); speech-to-text for the speaking parts. Daily limits per person, a
cache, and a monthly budget on the gateway keep it cheap.
- Recorded voices: every lesson and exam sentence is recorded once (Gemini TTS) and served
from merk-audio on GitHub Pages; anything else
uses the phone's German voice.
Darija
- Course: 20 lessons in two parts (A1.1, A1.2), each with the same four parts and a part
exam; no placement test or level exam yet. Syllabus in
data/darija/syllabus.md.
- Two scripts: every Darija text is shown in Arabic script with its Latin spelling
(Arabizi: 3, 7, 9, ch, kh, gh) under it; you type answers in Latin, and common spelling
variants are accepted. Spelling rules: data/darija/README.md.
- Flashcards: 608 words and phrases with French glosses (
pnpm import:darija).
- Voices and AI: everything is recorded (phones have no Darija voice); speech-to-text and
the AI teacher know Darija and answer in French.
- Lessons and words were written and reviewed by Claude, not yet by a native speaker.
Stack: Next.js, React, Neon Postgres, Drizzle, Better Auth, Vercel (AI Gateway).
Run your own
pnpm install
vercel link && vercel integration add neon # or set DATABASE_URL yourself
cp .env.example .env.local # fill BETTER_AUTH_SECRET and ALLOWED_EMAILS
pnpm db:migrate
pnpm import:words # German A1 → B2 + phrases, from data/ (backs up first)
pnpm import:darija # Darija A1 words and phrases
pnpm dev
AI feedback needs AI Gateway access: on Vercel the deployment's OIDC token is used
automatically; locally run vercel env pull .env.local (or set AI_GATEWAY_API_KEY). Set a
monthly budget for the project in the gateway. Without it, writing and speaking fall back to
marking yourself.
On iPhone: open the site in Safari → Share → Add to Home Screen, then log in inside the
installed app.
Content
Lessons, exams and word data are in data/ (see data/README.md for sources
and licences). After adding or changing lessons, run pnpm units:index; to record new
sentences, run scripts/gen-audio.ts (see its header) and push the merk-audio repo.
Keys
In reviews, Space shows the answer and 1–4 grade it (Again, Hard, Good, Easy).
Code is MIT.
