Range
In developmentA daily tone-guessing game, Wordle for vocal tone. You get a line and a target tone, record a two-second read, and an on-device emotion model scores how close you landed on a tone radar. One puzzle a day, a share card built to be self-deprecating, and zero servers.
The Idea
A daily game about how you say it.
Wordle owns letters. Range owns tone. The same line read sarcastic, flirty, disappointed, or supportive scores completely differently, and the gap between what you meant and what the model heard is the joke that makes it shareable.
Selected from 19 candidate concepts after four rounds of adversarial research, then gated behind real validation tests before any app code.
The Build
On-device, offline, no backend.
The emotion model runs entirely on the phone, so a read never leaves the device and there is no inference bill. Progression uses unlockable app icons that ship inside the binary with unlock state stored locally. The whole game runs with zero servers and zero per-play cost.
How a round works.
Prompt
A daily line plus a target tone to hit.
Record
A two-second read, captured on device.
Score
On-device emotion model plots your read on a tone radar against the target.
Share
A programmatic share card, designed to be funny when you miss.
Tech stack.
App
- Expo
- React Native
- TypeScript
- expo-alternate-app-icons
On-device ML
- Emotion / tone model
- Core ML conversion
- Offline inference
Tooling
- Python scoring harness
- Leave-one-out validation
- Programmatic share-card renderer
Status.
In active development. The on-device scoring harness is built and validated, the design and progression mechanics are locked, and the app build is in progress. This page will grow into a full case study as it ships.