Xelec hardware engineering
Firmware, boards, sensors and connected devices — engineered for production, not for the demo table. The same team that lets our AI run where the cloud cannot reach.
Capability
Most product programmes lose months in the handovers — design house to firmware contractor to manufacturer, each blaming the last. We hold the whole chain, so the trade-offs get made once.
Figures on this page are illustrative placeholders pending verified capability data.
Six services
Each engagement is scoped to a physical deliverable: a unit that runs, a design package a manufacturer can quote, a fleet you can update.
H-01
A connected product that works on the bench and falls over as a fleet.
End-to-end connected systems: device, gateway, connectivity and the telemetry pipeline behind them. Provisioning, identity and secure OTA are designed in at the start, because retrofitting them onto a deployed fleet is where most IoT programmes stall.
Deliverables
H-02
You need proof the thing can exist before anyone will fund it.
Concept to working prototype: dev-kit bring-up, breadboard rig, first custom PCB, and an honest feasibility read on power, thermal, size and unit cost. The output is a device you can put in front of a customer or a board, plus what it would cost at volume.
Deliverables
H-03
Firmware written once, then nobody can safely change it.
Bare-metal and RTOS firmware, board support and drivers, bootloaders with secure boot and signed updates, low-power state machines, and connectivity stacks. Written with a test harness and hardware-in-the-loop rig so the tenth revision is as safe as the first.
Deliverables
H-04
A prototype that no contract manufacturer will quote against.
Schematic capture, PCB layout, signal and power integrity, mechanical and enclosure coordination, design-for-manufacture review, test-fixture design and certification preparation. The deliverable is a documented design package a manufacturer can build from.
Deliverables
H-05
The model is blamed when the capture was wrong all along.
Vision, acoustic and vibration capture specified alongside the model that consumes it — optics, framing, lighting, sampling rate and synchronisation chosen for accuracy. Then the distilled model, quantised and validated on the target silicon under sustained thermal load.
Deliverables
H-06
A shipped device with no one left who understands it.
Board revisions, component obsolescence and second-source swaps, firmware maintenance, field diagnostics and fleet health reporting. Devices live for years; the engineering has to as well.
Deliverables
How an engagement runs
You can stop at any stage and walk away with the artefacts. Nothing is contingent on signing the next phase.
01
Constraints first: duty cycle, power budget, environment, certification target, volume and landed cost.
1–2 weeks
02
Dev-kit or first custom board, firmware skeleton, and a working unit that proves the risky part.
4–8 weeks
03
Layout hardening, DFM review, thermal and power validation, test fixtures and certification preparation.
6–12 weeks
04
A small production run in real conditions, instrumented, with OTA and diagnostics already live.
4–8 weeks
05
Transfer to manufacturing, then revisions, sourcing changes and fleet operation under an agreed SLA.
Ongoing
Why both under one roof
An AI company will send you a cloud API. A design house will send you a board. Neither owns the trade-off between accuracy, latency, power and unit cost — which is the only decision that matters once a device has to ship.
Capture designed for the model
Sensor choice, optics and sampling specified against the accuracy target, instead of the model being handed whatever the hardware happened to produce.
Compute sized to the workload
Silicon selected after the distilled model is profiled, so you are not paying for an accelerator you never saturate — or discovering you undersized it at pilot.
Data that never leaves the site
Inference at the station means the imagery, audio or documents stay on your network, and the audit trail is local. A residency answer that survives legal review.
One accountable team
When accuracy drops in the field, there is no argument about whether it is the model, the firmware or the sensor. Same team, same contract.
Next step
Cycle time, power budget, certification target, landed cost — tell us what the device has to survive and we will tell you whether it is a four-week prototype or a twelve-month programme.