01
GatesHub
Connected-data portal for the Gates Institute. Lab, business, and quality systems feeding one hub.
View
Operator · Scientist · Builder
Where Science Meets AI Strategy

Research operations, finance, and AI enablement at the Gates Institute, University of Colorado Anschutz Medical Campus.
I bridge cell-and-gene-therapy science and the systems that move it: finance, operations, and AI.
I am aggressively AI-forward about one thing: accelerating science. I ship agents and data systems into a regulated research institute, I teach the people around me to build their own, and I spent years in open and decentralized science arguing that the acceleration has to be shared. Personalized medicine is close. Who it reaches is still being decided.
My thesis
Two convictions, one from each half of my career. The science half taught me the first. The business half taught me the second. Everything I build sits where they meet.
Bench-to-bedside rarely fails at the bench.
It fails when the science is ready and the coalition is not. Payers, tech transfer, donors, and compliance all have to be bought in before the result arrives.
You earn that one room at a time: build the thing that solves someone's actual problem, let it work, and the next room starts the conversation already convinced. AI is what lets one operator run that across a whole institute.
And none of it should start from scratch. A postdoc leaves and the project should not die with them. A pilot ends and its proof should shorten the next one. Knowledge that compounds is the difference between a place that accelerates and one that keeps rediscovering what it already knew.

Not because you will write code.
Because agents do their best work in that shape: files, history, review, and an undo button.
Structure the work like software and an agent can actually help you run it. Leave it in inboxes and slide decks and it cannot.

Selected work
01
Connected-data portal for the Gates Institute. Lab, business, and quality systems feeding one hub.
View
02
Accreditation tooling that maps evidence to standards without overclaiming coverage.
View
03
Market intelligence on 120 cell-and-gene-therapy centers.
View
04
Personal knowledge assistant on an MCP server every Claude surface can call.
View
05
AI Pokémon TCG app in invite-only beta. Proof I ship outside the day job.
View
Also shipping · multimodal safety evals, ordering automation, procurement workflows. See automations →
Built to teach
Most business problems are not language problems. Six of the one hundred forty-five model approaches in the field are language models; the other one hundred thirty-nine are how you forecast demand, catch fraud, read an invoice, and route a ticket.

Now