- TypeScript 74%
- Svelte 23%
- JavaScript <1%
- Shell <1%
- CSS <1%
- 2 more languages <1%
awawa
A calm, predictable touch communication board built from a photo of a person's own world.
English · Español · 中文
Take a picture of a room. The AI finds the objects and outlines them; a grown-up confirms, and every object becomes a touch zone that speaks its word — in English, Spanish or Mandarin — with no connection needed afterwards.
It also works the other way round. In Listen, a grown-up says a sentence and the key words appear one by one, each with a pictogram, so the person can see what was said.
awawa is designed with neurodivergent people in mind — autistic people, people with ADHD, and others for whom speech is absent, unreliable or exhausting. It starts from a real photograph because a picture of your cup, your kitchen, your toys needs no symbol literacy.
Built during the Meta Global AI Developer Hackathon (Oct 26 – Nov 4, 2026).
Not a medical device, not therapy, not clinically validated. A communication-aid prototype, not yet tried with neurodivergent users.
Watch it work #
A narrated walkthrough in eleven short parts, from an empty install to a spoken phrase — profile, API keys, avatar, a first place outlined by AI, hand labelling, voices, offline, the camera, Listen and phrase building. The Demo guide has the step-by-step with screenshots. It exists in English, Español and 中文, each recorded in its own language.
| Part | What it shows | Summary | Subtitles |
|---|---|---|---|
| 01 · Profile | Name, language, sex | GIF | SRT |
| 02 · API keys | Masked Meta and DeepInfra keys, connection test | GIF | SRT |
| 03 · Avatar | Consent-gated avatar drawn by AI | GIF | SRT |
| 04 · First place | A photo → Meta finds and outlines the objects | GIF | SRT |
| 05 · Hand labelling | Rename, discard, draw an area the AI missed | GIF | SRT |
| 06 · Voices | New words get a voice; the place is published | GIF | SRT |
| 07 · Offline | Saved boards keep working with no network | GIF | SRT |
| 08 · Camera | Grown-up gate, one-button viewfinder, the photo becomes a board | GIF | SRT |
| 09 · Listen | Speech → transcript → key words → pictograms | GIF | SRT |
| 10 · Phrases on | The opt-in phrase strip | GIF | SRT |
| 11 · A phrase | Build and speak a phrase on a crowded board | GIF | SRT |
Why it is built this way #
| Principle | How awawa does it |
|---|---|
| Predictable | A touch zone never moves once saved — enforced in code, covered by tests |
| Concrete, not abstract | Zones are the outlines of real objects in a real photo — no symbol set to learn first |
| One touch, one word | No menus or levels; the phrase strip is opt-in and off by default |
| Low demand | No accounts, notifications, timers or streaks. The board waits |
| Calm | Dark, quiet interface; sound only on touch; honours "reduce motion"; bundled Atkinson Hyperlegible typeface |
| The adult curates | Nothing reaches the board until a grown-up confirms it — the AI proposes, never decides |
| Works when it matters | After a scene is saved, the board needs no network at all |
| Private by default | Keys, name, photos stay on the device; every send is consent-gated (Privacy) |
Read more #
- Demo guide — the full walkthrough, step by step, with screenshots
- Architecture — how a photo becomes a speaking board; models, storage, offline design
- Privacy — what leaves the device, what never does, and how consent works
- The mascot — why a rock hyrax without visible tusks
- Development — run it, test it, project layout, status and known limits
- SECURITY.md · ATTRIBUTIONS.md · LICENSE (Apache-2.0)
awawa is a communication tool. It does not diagnose, and it is not therapy.