AI Infrastructure & MLOps

AI Infrastructure & MLOps Consulting

Building and operating the infrastructure AI systems actually run on — from data pipelines to optimized, production-grade inference at scale.

A model is only as useful as the system around it. Much of my work is the unglamorous but decisive part: getting data flowing reliably, making training reproducible, and turning a research model into a service that's fast, observable and affordable to run. As a former Director of AI Engineering I've built these systems from scratch and led the teams that operate them — so I can help whether you need a second pair of hands or someone to set the technical direction.

What I help teams with

• MLOps & lifecycle — experiment tracking, reproducible training, model versioning, evaluation and CI/CD for models, so releases are boring in the best way.
• Data management & pipelines — ingestion, labeling workflows, dataset versioning and quality checks for image and other high-volume data.
• Inference optimization — reducing latency and cost through batching, quantization, model export (ONNX/TensorRT) and hardware-aware tuning, including edge devices such as NVIDIA Jetson.
• Cloud & deployment — packaging models as reliable, scalable services with sensible monitoring and rollback.

Background

I've built and run AI infrastructure in demanding production settings — leading AI engineering for digital pathology at Ultivue/Vizgen, and delivering real-time systems at Hensoldt where inference speed and embedded deployment were hard constraints, not afterthoughts. That mix of research depth and production experience is what I bring to infrastructure work. See more on the about section of the homepage.