NihonGoal!

Inspiration

Most Japanese learning apps begin with a fixed vocabulary list and a rigid daily streak. We wanted to make it more human, learning from the Japanese a learner meets in real life.

What it does

NihonGoal! is an AI powered Japanese learning companion designed for daily life. A learner chooses a support language and a starting comfort level, then can begin anywhere:

• Talk with Sensei, a conversational Japanese tutor
• Paste Japanese into Discover to turn something encountered in the world into learning flashcards
• Read a current news headline, then ask Sensei to rewrite at the learner’s level.
• Save, add, edit, and remove language in a personal library

How we built it

We chose Expo, React Native, TypeScript, and Expo Router so the first version could run in a browser while keeping a practical path to iOS and Android.

Challenges we ran into

The most interesting challenge was that “OpenAI-compatible” does not mean every provider behaves the same way. Different providers use different headers, completion-token parameters, browser CORS policies, and reasoning behaviour. We solved this by separating provider details inside the adapter, using JSON mode where it is supported, increasing output capacity only for the longer news-reading task, and accepting clearly labelled fields as a recovery format.

Accomplishments that we're proud of

• Turning an early product conversation into a real, working cross-platform prototype.
• Making the AI tutor useful
• Testing through a real browser with real provider responses

What we learned

We learned that the product design and the technical design cannot be separated in an AI-native app. We also learned that “serverless” is not the same as “offline,” and that it has useful boundaries. Most importantly, vibe coding worked best with HITL: the human supplied the vision, taste, priorities, and real-device feedback; the AI accelerated implementation, documentation, debugging, and testing.

What's next for NihonGoal!

• Add a proper global lookup and dictionary-pack experience
• Improve the adaptive learning model
• Expand source-linked reading with additional publishers
• Add optional speech input, TTS, and pronunciation feedback
• Improve photo discovery
• Add secure native credential storage,

Built With

• codex
• expo.io
• gpt
• node.js
• typescript
• web