All projects
Mobile · 2026
An English vocabulary app for Vietnamese learners: spaced repetition that knows when you're about to forget, and AI that fills the dictionary's gaps.
- Role
- Design & development
- Platform
- Android · Web
Overview
The problem
Generic flashcard apps don't know when you're about to forget a word. Free dictionaries rarely have Vietnamese meanings, real examples or usage notes, and some common words are missing entirely.
The solution
Vocabo ships about 3,300 curated words with English and Vietnamese meanings and CEFR levels. You review by typing the word, scheduled by FSRS, and when a word or sense is missing, AI builds it once and caches it for everyone.
Features
What it does
FSRS spaced repetition
Every card is scheduled by FSRS, with an adjustable retention target, maximum interval and daily limits.
Type the word, don't just flip
See the Vietnamese meaning and type the English word letter by letter; the grade comes from your attempts, not a self-rating.
A curated word store
About 3,300 frequent words to browse by level and frequency, with recorded or spoken pronunciation.
AI where the dictionary falls short
Missing words get a full entry, and each sense gets examples, collocations and usage notes, generated once and cached.
Progress you can see
A daily ring, streak, 30-day retention, level bars, an activity heatmap and your hardest words.
Android and the web
An Android build and an installable web app from one Expo codebase, with Vietnamese keyboard quirks handled.
How it works
From a missing word to a review
- 01
Sign in
A 6-digit email code or Google sign-in, so no deep links are needed on any device.
- 02
Look up
The app reads the shared word store straight from Postgres, protected by row-level security.
- 03
Fill gaps
Edge Functions keep the AI key on the server, build missing entries and cache them for every user.
- 04
Review
Each answer is logged and updates the FSRS card in one safe-to-retry database call.
Highlights
Engineering notes
A hard cap on AI spend
Daily limits are shared across all users, because per-user caps can be multiplied with new accounts; cache hits are free.
AI output you can trust
Structured output must echo the source word, the dictionary fallback runs only on a real miss, and every entry records its source.
One module, two runtimes
The same shared code runs in Node for seeding and in Deno for Edge Functions.
Tested at every layer
182 unit tests, 95 database assertions for permissions and reviews, and zero-cost edge tests with a fake key.
Under the hood
Stack & timeline
- Expo SDK 57
- React Native
- TypeScript
- TanStack Query
- Supabase
- PostgreSQL
- Deno Edge Functions
- OpenAI
- ts-fsrs
AI
Per-sense examples, collocations and usage notes, with a global cap on AI spend.
Review
Typed-answer review replaces flashcards, and the fox mascot arrives.
Android
496 more B1 words, an Android build, an installable web app and profile stats.
MVP
Word store, AI dictionary fallback, email-code sign-in and CEFR levels.
Gallery
A closer look
