I built a seven-day Korean speaking notebook with AI. Here is the app—and the build process.

I know a few Korean words. That does not mean the first sentence will come easily when I order in a café, ask a teacher to repeat something, or find the laundry room in a dorm. I am mio, a Japanese writer learning Korean and English. Those imagined moments before studying abroad gave me a smaller goal than “learn Korean”: practise one useful first line in each of seven everyday scenes.

I built a working browser app for that goal. This is one article: the finished interface, use cases and limits are free to read; the paid section explains how to reproduce and adapt the tool, including prompts, file placement, fictional sample data, tests, deployment and sharing.

The spark was a project, not a template

A developer’s Reddit post about building a Spanish-learning app with AI caught my eye. The resulting Spanish Buddy site describes 12 weeks and 84 lessons. Popularity is not proof of learning outcomes. What I borrowed was the design decision: make a small tool for situations the maker actually wants to handle. I did not copy that app or its code.

My seven scenes are a café, classroom, dorm, shop, transport, a future work meeting and a final review. These are imagined use cases—not events I claim to have experienced abroad or at work.

What the finished app actually does

Choose a day card, read the situation, speak once and write what you could say. Only then open a reference expression. Rate yourself “try again,” “almost” or “I could say it.” The progress display counts days in the last category. A session is designed for roughly three to five minutes; that is a design target, not a measured learning result.

The answer auto-saves in this browser. A Save button, JSON export and guarded import provide recovery options. There is no login, cloud sync, external AI API or automatic grading. Device speech synthesis can read a Korean reference if available, but pronunciation and voice vary. The Korean expressions are not yet native-speaker-reviewed; verify them with a teacher or reliable source before relying on them in real conversation.

How I would use it—and share it

On café day, I would ask for one cold drink to take away. Even if my first attempt is only a few Korean words, the attempt is mine. I can reveal the reference, notice one difference and repeat. On day seven I can revisit whichever scene felt hardest. This is a self-reflection tool, not a correctness score.

Share sends only the page URL and selected day. It does not put an answer, rating or name into the link. Each friend gets their own browser storage; our answers do not sync. Export records before switching devices or clearing browser data. Never write a name, address, school or confidential work information into a practice answer.

A localhost or file:// address works only on the maker’s computer. A friend needs a deployed HTTPS link. The screenshots here document local implementation and testing; public cross-device verification is a separate step, not a claim already completed.

What is in the paid build guide

The free takeaway: write down three moments where the first sentence matters, then try one line aloud before checking an example. If you want to build your own version, the paid section gives six reproduction prompts, a file map, fictional sample data, storage and privacy rules, a QA checklist, hosting and sharing steps, and ways to adapt the same app to interview or work practice. The eight source files are provided as collapsible, copyable code panels inside the paid section; they are image-free and contain no personal accounts or real learner records. The editorial portraits are not licensed for reuse.

Disclosure: The portraits are AI-generated editorial images based on mio’s approved source portrait, not documentary photos of a study session. Inspiration: the original developer post and Spanish Buddy; this app is independently designed.