sr_tpm · google_cloud · ml_infrastructure
Twenty years shipping ambitious, high‑stakes software programs at the world’s most demanding engineering organizations. Currently leading ML infrastructure reliability for GPU and TPU fleets at Google Cloud.
01 · about
I run the kind of programs most leaders quietly hope someone else takes — ambitious, cross‑org, technically dense, and visible from the top. At Google Cloud I lead ML infrastructure reliability programs for the GPU and TPU fleets that power the world’s most demanding AI workloads. Before that, I led AWS’s largest compute platform launches, shipped flagship Surface devices for Microsoft, and managed nine BlackBerry product introductions.
I hold an MBA, four U.S. patents, and a published engineering paper. The thread across all of it: pulling clarity out of ambiguity and turning it into something a team of 40+ engineers can actually build and ship.
02 · why ai
AI tools changed how I work — and that changed what I want to work on. I use Claude, Gemini, and ChatGPT daily, not as productivity shortcuts but as genuine thinking partners. Working on the infrastructure that makes these models possible, I’ve watched the technology move from impressive to transformative faster than any platform shift I’ve seen in twenty years.
What draws me to labs like Anthropic isn’t the hype. It’s the specific, hard problem: how do you ship the most capable AI systems ever built — safely, reliably, at speed, at scale? That is a program management problem. The stakes are higher, the cross‑org complexity is greater, and the cost of getting the design‑vs‑execution tradeoff wrong is more consequential than anything I’ve encountered before.
My career has been a series of platform‑defining bets — BlackBerry’s global supply chain, AWS compute at pandemic scale, Surface at $50M development budgets, and now ML infrastructure for Google’s GPU and TPU fleets. Frontier AI is the next platform. I want to help build the operational foundation that lets the best researchers in the world do their best work — safely.
I’ve started building with these tools too — a Python optimization add‑on for Google Sheets being the most recent example. Not because I’m becoming an engineer, but because the gap between “I have an idea” and “I have a working tool” has collapsed. That changes how I think about what’s possible inside a program.
03 · philosophy
Every program lives on a single dial: how much time you spend designing versus how much you spend executing.
Get that ratio wrong in either direction and the program fails — over‑designed programs miss the window, under‑designed ones burn cash rediscovering what a week of thinking would have surfaced.
The right ratio is not a constant. It moves with the cost of being wrong, the reversibility of the decision, and the rate at which the surrounding context is changing. A flagship hardware launch with a fixed factory window deserves more design up front than a software experiment whose first release teaches you more than any document could.
In AI development, this problem becomes asymmetric. Moving too slowly cedes the frontier. Moving without enough design produces reliability and safety failures that compound in ways that are hard to recover from. The dial matters more here than anywhere I’ve worked — and calibrating it correctly, in real time, at speed, is the whole job.
Everything else I do follows from that: how I structure milestones, how I escalate, how I trade scope, how I decide when to commit. Programs do not fail because people are unskilled. They fail because someone optimized the wrong side of that dial.
04 · experience
Senior Software Technical Program Manager — ML Infrastructure
Senior Supply Chain Product Manager — AWS Compute
Senior Technical Program Manager — Surface
Program Manager — NPI, Manufacturing, Supply Chain
NPD Engineer · General Manager
Created a $100M caps‑and‑closures product line the industry had pursued for two decades. Four U.S. patents and a peer‑reviewed engineering paper emerged from this chapter.
05 · selected impact
MTTR reduced
Led multi‑workstream program to lower Time to Repair for GPU/TPU machines through process latency reduction, improved fault attribution, and better diagnosability across the fleet.
google cloud · ml infraH100 program
Led cross‑functional program to reduce H100 tray swap rates and implement granular FRU schemas, delivering significant infrastructure Opex savings at Google Cloud scale.
google cloud · gpu reliability$900M
Built the case across product, finance, and SVP‑level stakeholders to commit advance purchases that pre‑empted pandemic‑era supply shocks across AWS global data centers.
aws · 2020–22First of kind
Launched AWS’s first Apple silicon cloud instance, solving an end‑to‑end supply chain that did not previously exist inside AWS.
aws · ec2 mac10,000+ hrs/yr
Architected and coded (PowerQuery) a BOM transfer process at Microsoft that cut cycle time from 2 weeks to 1 day. Now used by every Surface team.
microsoft · surface< 0.01%
Set the record for any Surface product at Surface Hub 2S launch — the lowest E&O ratio in program history on the most component‑dense Surface device ever built.
microsoft · hub 2s06 · skills
AI & daily tools
Data & analytics
Program leadership
Domains
07 · patents & publications
Four issued U.S. patents and a peer‑reviewed paper from an earlier chapter in engineering. They matter here because the program instinct is grounded in knowing how things actually get built.
US 20100278962 ↗
US 20080317898 ↗
US 20100272849 ↗
US 20100227019 ↗
08 · contact
Send a message directly, or connect on LinkedIn. I read everything.
kirkland, washington