Mirume.
Real-Time Japanese Reading Assistant
A macOS reading assistant that brings Japanese dictionary information, vocabulary levels, translation support, and review into the reading experience.
Tauri / Rust / React / TypeScript / FastAPI
Project motivation
Learning a new word shouldn’t mean losing your place.
Reading Japanese can involve frequent interruptions to look up unfamiliar words. Mirume explores how software can reduce that friction and keep learning close to the text.
Provisional motivation draft · to be reviewed by JimboStay in the story.
Mirume reads Japanese text under the cursor and provides dictionary information, JLPT vocabulary grading, and translation support. Spaced-repetition review helps carry new vocabulary beyond the current reading session.
Engineering &
implementation.
Built for the desktop
A Tauri/Rust and React/TypeScript overlay captures text with the macOS Accessibility API at 4.3 ms median latency. Apple Vision OCR provides a fallback at 520 ms median, with 99.4% character accuracy across 72 light/dark, multi-size test images.
Help in context
FastAPI serves JLPT-level word grading against 274k+ dictionary entries and 232k+ example sentences. Cached DeepL translations cut repeat lookups from 303 ms to under 10 μs.
From reading to review
Eighteen REST endpoints support SM-2 spaced-repetition review, connecting in-context reading with vocabulary practice.
Challenges &
decisions.
Add the verified text-capture implementation, a real performance or interface tradeoff, and how you evaluated reading accuracy.
A work in progress.
Add what you learned about desktop development and language-learning interfaces, plus your next planned improvement.