Offline-First Study Platform

The web application runs on React 19 and Vite. The mobile application is built with Expo and React Native and ships to the store through managed builds. They share the dictionary schema, the study-scheduling logic and the sync contract, so a definition corrected once is corrected everywhere.

Twenty versioned builds went out over the life of the engagement. Release was part of the work, not a phase that happened to someone else afterwards.

The first web implementation loaded the dictionary through a WebAssembly SQLite build in the browser. It worked, and it was slow: a heavy initial parse, a large resident memory footprint, and search that made users wait.

We migrated the storage layer to IndexedDB with indexes designed around the queries the app actually makes — prefix lookup, root lookup, category browse. The public API of the data service stayed identical, so the rest of the application did not change. A one-time background migration converts every entry on first launch behind a progress screen, and from then on the dictionary is local. Search became instant, and the 7.6 MB payload is fetched exactly once in the app’s lifetime.

Review scheduling uses FSRS, a modern spaced-repetition algorithm, rather than the fixed intervals most study apps still ship. Cards surface when the learner is about to forget them instead of on a calendar that ignores how they are actually doing.

Progress, study lists and statistics sync through Supabase, so a session started on a laptop continues on a phone.

Every core action — search, browse, review, record progress — completes with the network off. Sync is a background reconciliation rather than a precondition for using the product.

That constraint shaped the whole architecture, and it is the reason the app is usable on a commute, in a classroom with bad reception, or on a prepaid data plan.