// CASE STUDY 02 · EDTECH · AI
Fluent Air
Flashcards for language learners. Type a word and get everyday sentences at your level, each with audio and a picture. Anki decks come in with every review, and studying works offline.

- 15sentences per word, grouped by meaning
- 3Anki file formats imported with their full history
- 5interface languages
01THE CHALLENGE
Good flashcards use real sentences, with audio and a picture. Making one by hand takes minutes: find a sentence at your level, translate it, record or find audio, look for an image, then build the reverse card.
Many serious learners already use Anki and have years of reviews in it. A new app is only worth it if they can bring that history along and leave again whenever they want.
02MY ROLE
I built it alone, from the first idea to the live product: product decisions, interface, the AI pipeline, offline sync and the Anki importer.
- React
- TypeScript
- Vite
- Supabase
- Cloudflare Workers
- Dexie (IndexedDB)
- FSRS
- Gemini
- ElevenLabs
03KEY DECISIONS
- 01
The learner chooses, the AI drafts
The AI writes 15 sentences around the word, in three groups: everyday uses, other forms and tenses, and sentences that reuse words the learner finds hard. Nothing becomes a card until the learner picks it.
- 02
Offline first
Cards, audio and pictures are stored on the device. You can study on the metro without signal, and everything syncs between phone and computer when the connection comes back. No review is ever lost.
- 03
Bring Anki, keep everything
An importer written from scratch reads Anki files with their media, intervals and every past review, and the same FSRS scheduler keeps the learning curve where it was. Decks can be exported back to Anki at any time.
- 04
AI calls behind a server
Every AI request goes through a Cloudflare Worker that checks the user’s login and monthly allowance. Keys never reach the browser, and each provider has a backup so a busy service does not stop the app.
- 05
Built to move to native later
Scheduling, sync and the Anki parser live in a TypeScript package with no React and no browser code, ready to be reused in a native app.
04ON THE PHONE



05RESULTS
Live at fluentair.app, with cards made in seconds instead of minutes and an interface in five languages.
The app is in its soft launch. I will publish usage numbers here once there is enough real data to be honest about.
06WHAT’S NEXT
Learn from the first users, then decide on pricing and on a native app built on the same core package.
GOT A PROJECT
LIKE THIS?
Tell me what you want to build. I'll reply within 24 hours with straight answers and a fixed quote.
LET'S TALK →