AI & ML engineering
Edge AI partners, not another demo shop
On-device inference
Sensor-fusion models
Traceable by design
Cloud APIs and notebook demos fall apart once inference has to run on a battery-powered device, offline, inside a real-time loop.
Quantised, pruned models that run inside the memory, latency and power budget of the MCU or NPU already on your board.
We build the capture, labelling and drift-monitoring pipeline around your own devices, so the model sees the conditions it will actually meet.
Modular, documented code and a clear path to bring the work in-house when you're ready, no lock-in.
The sectors where our firmware, apps and data pipelines already run — and where on-device inference changes what the hardware can do.
Dive-profile models running on-device, offline, at depth
Real-time shot and movement analytics on ruggedized hardware
Inference at the edge, where private 5G and LoRaWAN links drop
Anomaly detection across RTLS and asset-tracking streams
Biometric and gesture models inside a wearable power budget
Capture, labelling and drift monitoring built as one pipeline
The network of brands and specialists Endcode builds extreme-environment technology with.
Dive equipment
5G connectivity
Survival & expeditions
Airsoft equipment
AEG customization
Hardware partner
Architectural detail for your engineering team.
PyTorch and TensorFlow for training and export; ONNX Runtime, TensorFlow Lite Micro and vendor NPU toolchains for deployment on MCU-class hardware. Sensor-fusion and DSP pipelines feed the model, so inference sits inside the same real-time loop as the firmware.
For models feeding IEC 62304 or ISO 13485 devices, and for systems in scope of the EU AI Act, we architect dataset provenance, versioned model artefacts and evaluation traceability as you go — not assembled retroactively before submission.
Modular, documented code with training scripts, dataset schemas and evaluation reports, plus an explicit transfer plan to your internal team written into the engagement from day one. Your data and your weights stay yours.
Blast Buddy's AI coach reads shot timing, transitions and accuracy off smart targets built on ultra-low-power microcontrollers, and returns coaching-level feedback with no human instructor in the loop.
Blast Buddy AI shooting coach — AROX AX-E0 smart targets
Every engagement ships evaluation reports against your own field data, versioned model artefacts, and the training pipeline that produced them — plus a knowledge-transfer plan to your internal team, written into the contract, not promised verbally.
