Sentinel — asset intelligence
Assets don't fail suddenly.
They drift.
Sentinel learns the behavioural signature of one specific yacht, vehicle, building or machine — then tells you the moment that signature changes. One self-installed device. No wiring. No installer.
The problem
Nobody is watching
A high-value asset spends 95–98% of its life unattended. The failure is the same in every market — only the cause differs.
Gulf — heat is the destroyer
Cabins exceed 70 °C
Batteries cook. Air conditioning fails unnoticed and mold blooms across an interior within days. Fuel vapour risk rises with temperature. Yachts and garaged supercars sit idle through the entire summer.
Northern Europe — cold and water
Frost bursts, damp rots
Pipes split in unheated cabins. Winter damp destroys interiors. Boats sink quietly at the berth and are found weeks later. Assets sit idle from October to May.
Why it stays unsolved. Wired hubs cost USD 600–1,500 plus an installer, take hours to fit, and stop reporting the instant power is cut — exactly when they matter. Consumer trackers report a position and nothing else. And fixed-threshold alarms on assets that move, heat and vibrate produce constant false alerts, so owners switch them off. Learning what normal means for each individual asset is a machine learning problem, not a sensor problem.
The intelligence
Four layers, one device
Signal becomes understanding in a defined sequence — each layer only sees what the one before it has already made sense of.
LAYER 01
Edge feature extraction
Physics-informed signal processing on the device turns raw motion, environmental and electrical streams into engineered features. Bytes leave the asset, not kilobytes — which is what makes cellular economics work at scale.
LAYER 02
Per-asset baselines
Unsupervised learning across roughly seventy signal dimensions. Every asset learns its own normal, because no two hulls, garages or machines behave alike. Absolute thresholds cannot work across a real fleet.
LAYER 03
Evidence fusion
Nothing fires from a single detector. Two physically independent channels must agree, with temporal persistence and contextual suppression, before an alarm reaches you. Target: under one false alarm per asset per year.
LAYER 04
Generative briefings
A grounded language model turns the week into a short, readable health briefing — strictly constrained to phrase measured facts, never to invent them, with automated verification on every output.
Every deployed device improves the models for every other. Detection quality rises with installed base — a compounding data advantage, trained across two of the most extreme operating climates on earth.
Capability
Inference, not wiring
Each of these is derived from physics the device can already observe — which is why there is nothing to install.
Water ingress
hull roll dynamics + trim shift
Theft & movement
learned position scatter + motion class
Power loss
charge-state signature — survives the cut
Mold risk
humidity / temperature dwell integral
Cooling failure
thermal response vs learned baseline
Battery health
voltage sag under known load events
Machine hours
vibration spectral signature
Fuel vapour
gas sensing, accumulation trend
Where we build
Abu Dhabi, on purpose
Two structural reasons, both of them engineering rather than convenience.
Hardest test environment on earth
Prove it here, it holds anywhere
Sealed electronics on an idle asset face four stressors at once: sustained 45–50 °C, coastal humidity near saturation, salt aerosol and airborne dust. No European location offers that combination. Validated through a Gulf summer means validated globally — and the reverse is not true.
Inverse seasonality
Two idle seasons, one model
Gulf assets sit idle through summer heat. Northern European assets sit idle through winter cold. Operating in both gives year-round exposure to the idle-asset condition across two opposite failure regimes — continuous training, and far better generalisation than either market alone.
Team
Two capabilities that rarely sit together
Teams from software underestimate the hardware. Teams from hardware underestimate that false alarms — not sensors — are what kill these products.
Robert Mevatne
Co-founder & CTO
Twenty years building certified rugged electronics for hazardous and extreme environments, including as R&D lead at BARTEC Pixavi, a world leader in Ex-certified industrial devices. Owns hardware architecture, embedded intelligence, environmental qualification and certification.
Qiuying Liu
Co-founder & CEO
Applied artificial intelligence and machine learning. Leads model architecture, the fleet learning pipeline, cloud inference and the generative communication layer, alongside commercial development and operations.
Contact
Get in touch
We're interested in conversations with asset owners and operators, marina and fleet partners, insurers, and investors.