Platform

The sensing platform behind everything we build.

Multi-modal sensors, on-device machine learning, and industrial integration — one architecture that adapts to any signal, in any environment. Stopyr is its first product; the platform is the engine.


The method

We read the pattern, not the threshold.

Conventional monitoring watches for a single value to cross a line. That's too late and too blunt — it misses the early build-up, and it can't tell a real event from noise.

AIOTOMATE analyzes the sequence and magnitude of sensor events over time. A rolling window of multi-channel sensor data is fed to a machine-learning model that recognizes the pattern of an emerging failure — the order in which signals appear, how they escalate, how they relate to each other — against patterns learned from recorded event data.

The output isn't a binary alarm. It's a probability of the event — graded, early, and continuously updated. That's what lets us warn before a threshold would ever trip, and distinguish a genuine event from an environmental false alarm.

Signal over time

Alarm threshold early-warning margin Pattern detected event forming Threshold fires event underway Time → Sensor signal

On-device intelligence

The decision runs on the edge.

Inference runs entirely on-device, on the STM32N6 NPU — no cloud round-trip, no connectivity dependency for the safety decision. A 1D convolutional model, accelerated on the NPU, processes the sensor window locally and outputs event probability in real time.

Models are trained offline on recorded event data using ST Edge AI tooling, then deployed to the device for inference. The intelligence lives where the sensors are — which means it keeps working during network loss, inside enclosures, and wherever a cloud round-trip is too slow to matter.


Gas
Thermal
Electrical
Vibration
Acoustic
Vision
+ any analog or digital signal

Extend to a new sensor

  1. Add a sensor
  2. Capture event data
  3. Train the model
  4. Deploy to the edge
Sensor-agnostic by design

Any signal. Any environment.

The method doesn't care what produces the data. Because it operates on time-series signals, the platform ingests any analog or digital sensor — gas, thermal, electrical, vibration, acoustic, vision, and more.

Stopyr applies this to a specific gas-sensor array for battery thermal-runaway detection. The same architecture extends to condition monitoring, visual inspection, and process safety — the model changes, the platform doesn't.


Integration

Works with what's already installed.

AIOTOMATE speaks the protocols industrial sites already run — RS485 / Modbus, CAN, and dry-contact I/O into existing BMS, SCADA, fire-and-gas panels, inverters, and access control. Manufacturer-agnostic, retrofit-ready, no rip-and-replace.

Modbus RTU/TCP

RS485 industrial fieldbus

CAN bus

Battery & inverter comms

BACnet · BMS

Building & EMS integration

Fire & gas panels

Relay · dry-contact I/O

PCS · inverters

Hybrid & storage inverters

MQTT · REST

SCADA · cloud edge


Visibility · Dashboard, API, Cloud

Local decisions, full-fleet visibility.

The edge makes the decision; the platform gives you the picture — across every site.

Management dashboard

Device provisioning, configuration, live monitoring, and reporting across sites — one pane for the whole fleet.

Remote API

Data and status exposed for integration into your own systems — pull results into the tools your operators already use.

Cloud visibility

Results stream to the cloud for cross-site trends — an optional visibility layer, never a dependency for the alert.


Where it stands

Proven approach, edge-native platform.

Honest status, layer by layer — what's in testing, what's proven, what's operational today, and what's in build. Nothing overclaimed.

In testing

Detection approach tested against real battery packs — targeting BESS deployments from residential to grid-scale.

Proven

Hardware integration verified on the current generation in lab conditions.

Operational

The management platform and API are operational.

In production

Edge vision built on the same principles runs in production with a UK manufacturing partner.

In development

The STM32N6 edge platform is in active development — bringing the full inference stack on-device as our core product.