Every app starts with a question.
How we decide what to build, how we test it with families, and what we refuse to measure.
Three rules we work by.
They came out of building Lighthouse AAC and they govern everything we make.
Build with families first
Every project opens with time in homes and clinics, before a line of product code exists.
Measure on paper, not by telemetry
Our apps send nothing home. Outcomes come from families and clinicians who opt in under ethics review and informed consent, never silently from the device.
Publish the instruments
Protocols, questionnaires and scoring sheets are shared on request so results can be challenged and reused.
Four signals, one quiet decision.
Predictive Glow raises the contrast of likely words. It never moves a button, and it runs entirely on the device.
Time of day
Morning vocabulary surfaces in the morning. Bedtime vocabulary at bedtime.
Place context
Home and school each have their own most likely words. The place is recognised from a one-way hash of the Wi-Fi network name, computed on the device and never sent anywhere.
Sequential history
“Want” is usually followed by a noun. The app weights what tends to come next.
Recency
What the child just said matters more than what they said last week.
Nothing about your child leaves the tablet.
No accounts, no analytics, no advertising SDKs. The prediction model is a small function running on the device. This is a constraint on the whole foundation, not a feature of one product.
The app makes exactly two network requests, and only when a parent chooses to: checking for updated words and pictures, and sending feedback. Both are documented line by line.
Read the app privacy policy- Accounts required
- None
- Analytics or tracking in the app
- None
- Advertising SDKs
- None
- Where the model runs
- On the device
- Network requests
- Only when a parent taps