Skip to content

AI & security

Edge & Embedded AI

Some decisions can't wait for the cloud. A defect on the production line, a fault in the field, a mission with no connection. Edge AI takes the model to where the data is born. Because we design the circuit board, the embedded software and the model in the same team, we can meet latency, power and cost targets together.

What we do

Hardware selection and design

We choose the processing power the task calls for, and design the board and its thermal management for NVIDIA Jetson, NPU-equipped application processors or microcontrollers.

Model optimisation

We shrink models through quantisation, pruning and distillation, and fit them onto the target hardware while measuring the loss in accuracy.

Vision and audio processing

We run tasks such as object detection, classification, anomaly detection and keyword spotting in real time.

Sensor fusion

We combine vibration, current, temperature and position data to pick up the early signs of failure on the device itself.

Field updates and monitoring

We update models and software remotely and securely, and monitor their accuracy in the field.

Data collection and labelling

We design the device and the process for collecting the right data, and set up the workflow for labelling and data quality.

Typical deliverables

  • Hardware and model architecture proposal
  • Optimised model and accuracy report
  • On-device inference software
  • Latency, power and accuracy measurement report
  • Field update and monitoring infrastructure
  • Data collection and labelling process

Tools and platforms

  • NVIDIA Jetson
  • TensorRT
  • ONNX Runtime
  • TensorFlow Lite Micro
  • ExecuTorch
  • OpenVINO
  • PyTorch
  • OpenCV
  • STM32
  • NXP i.MX

Standards and methods

  • IEC 62443-4-2
  • IEC 61508
  • ISO 26262
  • ISO/IEC 42001
  • EU AI Act

Applied as the project requires.

Our approach

Success in edge AI is not about finding the best model. It is about finding the best model that fits the constraints. Latency, power consumption, cost and accuracy all affect one another, so we never decide on the hardware without the model, or on the model without the hardware.

We try the prototype early, with the real sensor and in the real environment. A model that does very well in the lab can behave differently in the light and noise of the field.

Where we apply it

Defence & Aerospace

Electronics, embedded software and mechanics designed for harsh environments, long product life and rigorous documentation.

Industrial Automation

Control boards, embedded software and operator interfaces that run on the production line for years without interruption.

Automotive

Vehicle electronics, embedded software and test systems, built with functional safety and cybersecurity in mind.

Engineering it draws on

Electronics & PCB Design

From schematic to a board ready for production. Power, signal integrity and testability are part of the design from day one.

Embedded Systems

Software that lets hardware speak. From microcontrollers to embedded Linux; secure, and updatable in the field.

AI & security

More AI & security services

AI Agents & Enterprise Assistants

AI assistants that work with your organisation's documents and systems, show their sources and deliver measurable results.

  • Software
  • Systems

AI-Assisted Legacy Modernisation

We analyse legacy systems with AI-assisted tools, safeguard them with tests and move them, step by step, to a modern architecture.

  • Software
  • Systems

Secure-by-Design & Product Cybersecurity

We build connected products to be secure from the design stage, and prepare them for the European Union's new cybersecurity rules.

  • Systems
  • Electronics
  • Embedded
  • Software

Let's talk about your project.

Describe it in a few sentences and we'll find the right starting point together.