Portfolio

Selected work, built and shipped.

Client projects and industry systems across machine learning, robotics, software, and hardware design. Each one started with a hard problem and ended with something running in production. Looking for research or personal projects? See Research and Projects.

40+

Systems Shipped

15+

Years of Experience

18+

Client Organizations

All WorkMachine LearningRoboticsSoftwareProduct DesignResearchMLOps

Machine Learning·2024

Aegis — Defect Detection for Smart Manufacturing

A deep learning pipeline that classifies surface defects on automotive components in under 40ms per frame. Trained on a proprietary dataset of 180,000 labelled images using a custom EfficientDet backbone. Deployed on NVIDIA Jetson AGX Orin edge units across three production lines in Germany. Cut manual inspection time by 72% and false-reject rate by 38% within the first 90 days of deployment.

EfficientDetEdge DeploymentNVIDIA JetsonMLOpsIndustrial CV

Robotics·2023

Helios — 6-DOF Pick-and-Place Controller

RL policy for a collaborative robot arm trained in IsaacGym, transferred to hardware. 94% grasp success across 30+ object geometries.

Hardware·2024

Lightweight Drone Frame — FEA Optimisation

Carbon-composite quadcopter frame, CAD-designed and FEA-validated. 31% lighter at equal stiffness, produced via SLS printing.

Software·2022

Kepler — Open-Source Motion Planner

Modular C++ motion planning library with Python bindings and ROS 2 integration. Used by three research groups, cited in 12 papers.

Machine Learning·2025

Atlas Energy Demand Predictor

Temporal Fusion Transformer-based forecasting pipeline for grid-scale energy management. 18% MAPE reduction over baseline model.

Robotics·2025

Pulse Perception Stack for L4 AVs

Multi-modal sensor fusion pipeline for adverse weather autonomous driving. 28ms inference, +41% rainy-weather mAP over previous model.

MLOps·2024

Orion Model Monitoring Platform

End-to-end MLOps platform for drift detection, retraining triggers, and explainability dashboards. Deployed across 12 production ML models.

Hardware·2024

Bionic Prosthetic Socket

Patient-specific prosthetic socket designed using parametric CAD and topology optimisation. Produced via selective laser sintering in PA-12 nylon.

Research·2024

Mira Quadruped Gait Policy

Sample-efficient RL policy for quadruped locomotion over unstructured terrain. ETH Zürich collaboration, published at IEEE ICRA 2024.

How I Work

From first conversation to shipped system

Every engagement follows the same process: enough structure to keep things on track, and enough flexibility to adapt to what the problem needs.

Discovery & Scoping

We start with a free 30-minute call, followed by a deeper dive into the problem. I ask the uncomfortable questions, audit what you already have, and define a realistic technical plan with clear deliverables and success criteria.

Research & Prototyping

Before any production code, I build a minimal prototype to de-risk the core technical assumptions. This phase usually surfaces the real problem, which is often different from the stated one.

Build & Evaluate

Clean, documented, testable code with weekly check-ins and continuous evaluation against the agreed success criteria. I don't disappear into a build phase and resurface with a finished product.

Handoff & Support

Every engagement ends with a full handoff: code, documentation, an evaluation report, and four weeks of post-launch support. Your team should understand what they have and be able to maintain it.

Ready to start a project?

I take on a limited number of engagements each quarter so that each one gets real attention. If you have a problem worth solving, I'd like to hear about it.

Start a conversation Read client stories