I architect solutions to operational problems in financial services — and I know how much automation each one actually warrants.
Twenty-one years turning high-stakes, manual financial workflows into systems that run themselves. I work with development architects and product owners to get the design right — and I came up through the code, so right means buildable. Lately: where agentic AI belongs, and where it doesn't.
Give me a problem where the current answer is “that's just how it works” or “we'll buy a tool for it.” I'll show you the solution the situation actually calls for — usually smaller, cheaper, and closer to what you already own than the room expects.
Architecture, not task lists
I design the solution end to end — the process, the data flow, the integration path — then work hand-in-glove with the architects and product owners who build it.
Grounded in what's buildable
Two decades of engineering sit under the analysis. My designs don't fall apart on contact with the codebase, I can tell when a vendor demo is hiding the hard part, and when a system is slow I find the real reason — not the one everyone assumed.
Automation, measured
Agentic AI is a tool, not a mandate. The questions that matter: where it fits, how to wire it in safely, and how much of a workflow it should touch before you're adding fragility instead of removing toil. I build these systems myself, so it isn't theoretical.
Minimum viable, elegant, on infrastructure you already own. Every engagement ends with a number that answers one question — what changed because I was here.
Middle Office ran wires by hand; the default answer was a vendor tool. My analysis showed the firm already owned the rails — the platform could be built in house. 0% to 99.9% straight-through processing, 99.9% instructed via SWIFT, a multi-million-dollar tool avoided. The client’s treasury team took Highly Commended for Top Treasury Team at the Adam Smith Awards 2024, in part for the programme. Read the write-up ↗
Not once. Client onboarding at Citigroup went from a five-day slog to near-instant. Corporate-actions settlement at BNY Mellon ran straight through on SWIFT. Same move each time: find where the process breaks, design the break out of it.
A bankruptcy-suspense payment process at SunTrust, automated — 50 person-hours a day returned, $29k a month saved. The Nevada-loan escrow calculation went the same way. Small, unglamorous, compounding.
Claude-powered agents that read fund inception documents and drive LemonEdge onboarding — weeks of manual setup per fund, gone. Backed by a personal stack of production systems I design and run myself.
I started in 2005 writing C++ for settlement systems at Goldman Sachs, then years of PL/SQL and data work at Northern Trust, BNY Mellon, and Citigroup — SWIFT straight-through processing, the technology side of Private Bank client onboarding. Working close to the users, I saw the leverage wasn't in writing more code — it was in getting the solution right before anyone wrote any. So that's where I moved. SunTrust, PNC, and a $50B PE firm since.
Goldman Sachs · Northern Trust · BNY Mellon · Citigroup · SunTrust · PNC · PE firm ($50B)
Full history on LinkedIn ↗
I work best with PE firms, fund administrators, and fintech platforms that need someone who can hold the business problem, the data architecture, and the automation path in one head. Based in Pittsburgh, remote with teams anywhere. I build production-grade systems on my own infrastructure for the joy of it — see the projects.