Embedded & Edge AI
Models that actually run on the hardware you are shipping.
Most machine learning works fine on a workstation and falls apart on a battery-powered board. I take models the rest of the way: choosing an architecture that fits the silicon, quantising it until it meets the latency and power budget, and building the inference pipeline around it so it survives contact with real hardware.
Typical engagements
- Get an existing model running on a target board within a power and latency budget
- Architecture and quantisation review before you commit to hardware
- Build an on-device inference pipeline with sensor input and real-time output
- Stand up orchestration and OTA updates across a fleet of edge devices
- Second opinion on whether an edge deployment is feasible at all
Proof
Work that demonstrates this capability in practice.
Reconfigurable CubeSat Clusters
A compute and communications stack that turns a swarm of 1U CubeSats into a single elastic cluster, with FPGA payloads that change function in orbit and workloads that migrate between satellites in seconds.
Read the case study →NASA MINDS 2025 - Exceptional Experimental DesignAir-and-Water Autonomous Vehicle
AQUAD: a single vehicle that flies to a site, submerges, manoeuvres underwater and returns to the air, on one charge. Built against a brief targeting ocean worlds such as Europa.
Read the case study →NASA MINDS 2024 - FinalistAutonomous Aerial Docking and Recharge
A heavy-lift carrier UAV acting as an airborne charging hub and data relay, letting micro-drones dock in flight to recharge, offload sensor data and borrow GPU cycles.
Read the case study →US Navy research programmeNatural-Language Flight Control
A chat-to-fly interface for UAV operators: spoken or typed instructions translated directly into flight commands by a language model running on the aircraft, not in a datacentre.
Read the case study →