Selected work
From research to deployed systems
Bluestar Software · Machine Learning Scientist · 2019–2023 (KTP, then staff)
Automated footwear identification — paper, patent, national service
A deep learning system that embeds footwear-outsole images and retrieves matches from a footwear database by nearest-neighbour search over a k-d tree of embeddings, with augmentation for partial or obscured crime-scene marks. Granted as UK patent GB2622032 (2024; proprietor Bluestar Software); I am first-named inventor. The underlying method — multi-label convolutional descriptors with transfer learning from low-resolution greyscale impressions — is published in Applied Soft Computing.
The matching technology is used in the UK National Footwear Service, operated on behalf of the UK Home Office. The service runs across 32 of the 43 police forces in England and Wales, handling around 36,000 queries a day at sub-second response times. Alongside the footwear programme, I led large language model work for legal and forensic text analytics, built speech-to-text acoustic models, and computer-vision re-identification for criminal-justice systems.
Metrolla · Lead Perception Engineer · May 2026–present
Edge LiDAR perception for people and vehicle analytics
I lead perception development for Metrolla's edge LiDAR platform in ROS2: sensor integration, 3D object detection, multi-object tracking, and occupancy and flow analytics. I brought new LiDAR hardware into the pipeline end to end — from raw point cloud to working detection and tracking — extended the platform across additional sensor models behind a common interface, and redesigned core tracking and scene reasoning from clustering-based methods to a probabilistic approach.
Cron AI · Technical Lead, Machine Learning · May 2023–Dec 2025
senseEDGE — real-time 3D perception at the edge
I led the machine learning function for senseEDGE, an edge-native 3D perception platform that turns raw LiDAR point clouds into real-time object detection, classification and tracking. I owned the model lifecycle — training, evaluation and release — and ran neural architecture searches and feature development under real-time, on-device constraints; delivered remotely with a London-headquartered team.
Bournemouth University · Doctoral research · 2016–2020
Physics-inspired link prediction in networks
My doctoral work reframed link prediction through Newton's law of universal gravitation: network centrality as mass, dissimilarity as distance, attraction as the likelihood of a connection. Published in Physica A (vol. 523) and since cited in graph neural network research. Related work includes NetSim, a framework for complex-network generation (Procedia Computer Science), and I worked with graph convolutional neural networks during the PhD.
Invento Software Limited · Deep Learning & AI Consultant · Feb 2026–present
AI-native enterprise systems
Brought in to lead Invento's transition from enterprise software delivery to AI-native systems: embedding predictive intelligence into ERP platforms for forecasting and optimisation, adding intelligent decision support and automation to CRM and business platforms, and designing AI-native architectures that integrate with the company's existing SaaS products.
