I designed and built OVERSTEP, a cross-platform AI language-learning app that generates practice on demand for any language a learner asks for — from Japanese and Chinese to endangered languages like Rapa Nui — and evaluates every answer with an explanation instead of a simple right/wrong. It is live on the Apple App Store.
OVERSTEP — AI-Driven Language Learning App
Most language apps ship a fixed bank of hand-authored exercises. OVERSTEP doesn’t. It uses large language models to generate exercises, grade open-ended responses, and explain why an answer is right or wrong — then adapts the difficulty to each learner’s performance. The result is a learning system that scales to languages no commercial app would ever bother to support.
- Generative practice for any language:
- LLM-generated exercises across vocabulary, grammar, and conversation
- Works for mainstream languages (Japanese, Chinese) and long-tail / endangered ones (e.g. Rapa Nui)
- Multilingual hints and explanations — a learner can study Japanese with prompts in their own native language
- Automated evaluation with feedback:
- Open-ended answers are assessed by the model, not matched against a fixed key
- Every attempt returns a clear, specific explanation of the reasoning
- Per-item Learning Level and Complexity scoring drives adaptive difficulty
- Progress, streaks, and community:
- Personal progress dashboard — completions, accuracy, daily streaks, category coverage
- Full practice history with the learner’s own answers preserved for review
- A community layer and daily tips to sustain the habit
- Production engineering & responsible AI:
- React Native (Expo) front end; Node.js / Express back end; MongoDB Atlas
- Deployed to production on Vercel + Heroku with analytics and error monitoring (Firebase, Sentry)
- Structured and stored user-interaction data — responses, performance, behavior logs — to power adaptation
- Deliberate responsible-AI decisions on user data, privacy, and safe handling of model inputs and outputs
OVERSTEP took Second Place at the Brightlands Startup Challenge (2024). Below are a few screens from the live app.
