The platform
Privacy isn't a policy here. It's a consequence of how the system is built.
HeliTrust is a federated-learning platform: every device computes its own Unifying Contextual Weights on its own motion data, and only ever sends home a small, anonymous set of statistical updates. No raw sensor reading, location, or health record is ever transmitted or stored centrally.
The way this is usually built
- • Raw accelerometer, GPS, and health-adjacent data leave the device on every reading
- • Privacy is a clause in a data-processing agreement, not a property of the system
- • A server breach exposes real health and location records
- • Data usually lands outside the EU, inheriting cross-border transfer risk
How HeliTrust is built
- • Each phone trains on its own motion data locally, in place
- • Unifying Contextual Weights evaluate motion telemetry on-device without sharing raw data
- • A compromised server yields model weights and hashed credentials, never a sensor reading
- • Self-hosted on-premises or in an EU data center, no transatlantic transfer to reason about
How it works
Three components. Nothing sensitive ever leaves the first one.
No pendant, no wearable, no proprietary hardware to distribute and replace. The phone someone already carries is the sensor; the rest of the system exists to notice, escalate, and prove what happened.
The monitored device
Any Android phone from 2018 onward, installed from a simple download page that detects the platform automatically. Accelerometer and gyroscope stream at 50Hz through a foreground service that keeps working with the screen off, after the app is killed, and after a reboot.
- HTTP POST every 1 second, per-device enrolment token
- Unifying Contextual Weights computed locally
- Survives app kill and reboot
The HeliTrust server
Receives a telemetry post, checks the key, aggregates Unifying Contextual Weights against a personal baseline, and computes a risk score without raw sensor readings ever leaving the device.
- Weighted FedAvg + secure aggregation
- EWMA baseline + fatigue signal
- 10-year AI Act audit log
Operator dashboard & family portal
The screen a caregiver or safety officer actually watches, re-skinned per domain, plus a bcrypt-hashed PIN status page for family; no app install, no account, just a link.
- Live per-person risk charts
- Four roles: admin / manager / caregiver / viewer, scoped to one facility
- Installs as a PWA on any device
Engineering maturity
Built like infrastructure, tested like one.
18
automated test files, run in CI, covering auth, tenancy, and anomaly detection
4
account roles, each scoped to exactly one facility
23
database tables, authenticated by default
€0
hardware cost, runs on the phone already owned
Want the technical deep dive?
A full architecture walkthrough under NDA is available for pilot partners and investors.
Request a pilot