About

I am a principal-level architect for AI factories and data centres, based in Sydney. For twenty-five years my job has come down to one decision made over and over in different forms: where should the compute physically live, and what has to be true for it to survive there. What makes it work is range. I am equally at home defending a multi-year investment case to a steering committee and elbow-deep in the fabric and the silicon, and most of the value is in translating faithfully between those two rooms. Boardroom to bare metal, and back. The story page tells that arc in order. This is the shorter version, and the part the CV leaves out.

It started somewhere unglamorous. My first data halls sat inside heavy industry, next to refineries and on mining platforms, in hazardous zones where a wrong assumption is a safety incident rather than a bug ticket. You learn fast in a place like that. You learn to start from the physics and the constraint, not from a reference diagram.

From there the scale grew but the question stayed the same. I spent years building liquid-cooled GPU farms, immersion and direct-to-chip, with the power distribution and the low-latency systems to match, before anyone was calling that "AI infrastructure." Then a long chapter moving whole regulated enterprises to the cloud, landing zones and guardrails and the 6 Rs, the multi-million-dollar programs where a bad migration surfaces weeks later and someone senior asks why. That is where the engineer learned to run a program end to end.

Today I do this for a Tier 1 Australian telco as the enterprise storage subject-matter expert for AI and HPC, and as a senior solution designer across on-prem, hybrid cloud and edge. Fabric topology, parallel storage, the air-to-liquid transition, and the governance over space, power, cooling and redundancy that decides whether an AI factory actually works. Much of the work lives at architecture-governance altitude: design assurance, reference patterns and landing-zone blueprints, guardrails as code, the vocabulary of TOGAF and the Well-Architected pillars, and the compliance line that regulated shops hold against APRA CPS 234 and ISO 27001.

The other end of that range is commercial. Architecture that cannot show its worth is just expense, so I speak the business case as fluently as the design: ROI, TCO, FinOps and program economics. I led a review of more than a thousand applications that returned north of twenty million dollars a year, ran the 6 Rs on economics rather than dogma, carried multi-million-dollar programs as principal and technical lead, and kept FinOps in the migrations so the move did not simply relocate the waste. Then I take the same decision down to the low-level design and the blueprint and make sure the thing we sold is the thing we can build. I hold the NVIDIA AI infrastructure certifications too, a formal edge on instincts I built the long way. The hardware changed at every step. The thinking did not.

Somewhere along the way I started building things at night. Not because I had to. Because I would see a problem during the day, think about it on the drive home, and by midnight have a prototype running. A warehouse platform because a friend's business was drowning in spreadsheets. A price tracker because I got tired of overpaying for camera gear. A self-hosted gateway to drive my model sessions from anywhere, because I was tired of an agent dying the moment my laptop slept.

I don't just call APIs. I own the silicon. I run a self-hosted NVIDIA DGX Spark and a cluster of Mac Studios, loading, quantising and evaluating local models, building evaluation datasets, and staying close to the training and fine-tuning tradeoffs, so I can see where these machines actually strain before I design a platform around them. When a model gets something wrong I want to know if it is the data, the quantisation, or the prompt, because the fix is different for each.

Around the accelerators sits a scaled hybrid lab: containers, orchestration clusters, multiple cloud regions, RAG and shared vector memory, and self-hosted multi-agent platforms, all meshed and secured end to end. It is where I test architectures before I recommend them to anyone. If I am going to stand in a steering committee and say something works, I want to have broken it first. Everything I make, I use.

I don't really have a specialty. I have a pattern. See a problem. Understand it deeply. Build a system that solves it. Then build an agent that runs the system. The domain changes. The approach doesn't.

I still get genuinely excited when things click. When a small model running locally extracts business rules from a photo of a receipt. When a single Go binary embeds an entire mobile web app. When a Telegram bot renders an AI agent's permission dialog as inline keyboards and you approve a code commit from the bus. Those moments are why I still build things after twenty-five years.