Distill — notes that teach you back, entirely on-device

Notes that teach you back

Role: Solo — design, build and release
Period: 2026 — in development
Status: In development

Distill is a privacy-first learning companion. A note is the only thing you create — write or paste something you want to learn and the app tidies it into clean markdown, then builds study material from it. The model runs locally, so nothing is uploaded.

Highlights

On-device inference — no server, no account
Notes are the only input; everything else is generated
Apple HIG on iOS, Material 3 on Android

Expo
React Native
On-device LLM
llama.cpp
SQLite

The idea

Study apps make you build the study material. Distill inverts that: you write one note, and lessons, flashcards and quizzes fall out of it. No decks to maintain, no import step, no second app.

The hard part

Running a useful language model on a phone is the whole engineering problem. The constraint that decided the architecture was context length — a study note is long, and a 2K-token window is not enough to work with one, which ruled out the obvious on-device runtimes and pushed the app onto llama.cpp.

Design

Both platforms get their own design language rather than one cross-platform compromise — Apple HIG on iOS, Material 3 on Android. It costs more work and it is the difference between an app that feels native and one that feels ported.

Stack

Expo + TypeScript, React Native
On-device LLM inference through llama.cpp
Local SQLite; notes are the source of truth
Platform-native design systems on each OS

Status

In development. Not shipped yet.