Sweet PalLanguage
From free text to structured data.
“Pasta with tomato sauce, a normal plate, bolus five minutes before” becomes a meal with a time, foods, estimated grams and a linked bolus. Plate photos go through on-device recognition.
An AI that says “not yet” when it doesn't know.
Sweet Pal pairs a language model with a statistical engine that works on each user's own data. The first understands people; the second keeps things grounded. What it notices stays a clue until it becomes a pattern.
Coming to the App StoreFor iPhone. We're running the final checks.
The app currently speaks Italian.
How it reasons.
A type 1 diabetes app is a hard test bed: the data is noisy, people are tired, mistakes are costly.
House rules.
Product choices before technology choices.
- No invented curves.
- Without a sensor, no line is drawn between two readings: that would be data that doesn't exist.
- No dosing advice.
- The model is instructed never to suggest insulin units.
- Explicit consent.
- Data analysis runs only if the user opts in, and switches off in one tap.
- Native to iPhone.
- SwiftUI, Apple Health, Liquid Glass. No web wrapper.
Inside the app.
Notes
What keeps coming back, and what quietly changes.
1 / 10
Watch it work.
Sweet Pal is a native iPhone app.
Coming to the App StoreFor iPhone. We're running the final checks.