work / mandarin-tutor · Built
Mandarin Tutor
Real-time voice lessons that feed a spaced-repetition word bank.
Week of Claude Maxing
Play with it
A pure-frontend replica with mock data: nothing is saved, nothing leaves the page. The real app is richer; this is the feel of it.
loading the demo…
The problem
Speaking practice and vocabulary memory live in different apps, so neither compounds.
What it does
Live voice conversation lessons (ElevenLabs) that debrief into an FSRS spaced-repetition word bank, with tone-coloured pinyin, tap-to-hear audio and a built-in HSK curriculum.
The thinking
Target user
Intermediate Mandarin learners (I am one, HSK 4 and up) who can study vocabulary but rarely get to speak.
Pain points
- Speaking practice and vocabulary memory live in separate apps, so neither compounds
- Tutors are expensive and hard to schedule
- Flashcards without audio teach reading, not conversation
Gap in the market
Real-time voice conversation that debriefs into a spaced-repetition word bank, so every lesson feeds tomorrow’s reviews automatically.
AI + tooling
- ElevenLabs conversational voice
- Claude (debrief + sentences + mnemonics)
- ts-fsrs spaced repetition
- Claude Code (build)
Frameworks
- Spaced repetition against the forgetting curve (FSRS scheduling)
- Comprehensible input: conversation slightly above current level
- Dual coding: tone-coloured pinyin + audio pairs visual and sound memory
- HSK 1 to 6 curriculum strands
- Zero-key core loop (the paid layer is an upgrade, not a dependency)
Success metrics
- TBD: words retained, lessons per week, review streak
Design the core loop to work with zero API keys; the paid layer is an upgrade, not a dependency.