Backend · AI systems · Full-stack
Trading engines, exchange integrations, and the agent systems teams are pushing to production without evals. Ten years of building where downtime is measured in dollars per second — most recently a hedging engine holding exposure for a $38B exchange.
[01]Selected work
Contract senior engineer
A hedging engine holding market exposure for a $38B exchange.
Contract
PRDs in, reviewed test plans out. 60% of the cases run without a human.
Contract tech lead
$6B+ routed across 8+ exchanges, on AWS ECS.
[02]What I do
The demo worked. Month three is the problem — silent regressions, a bill nobody forecast, and no way to tell whether a change made it better or worse. Agent architecture, evals that catch regressions, retrieval and tool design, cost and latency control.
One upstream goes down mid-request and takes your endpoint with it. High-throughput services and the integration layer under them — many unreliable providers behind one internal contract, with the failure handling for when one dies. Deployed on your infrastructure, not a prototype.
Architecture through deployed-and-working, without supervision. Useful when the plan is agreed, the quarter is committed, and the one engineer who could own it already has a job.
[03]How it works
Thirty minutes. Bring what is breaking. You get a diagnosis and a scope, or an honest no.
What I will build, what I will not, how long, what it costs. Priced per engagement — no packages, because your problem is not a package.
Shared channel, weekly demo, every commit visible. Runbook and walkthrough are deliverables, not favours.
Bring the thing that’s breaking. You’ll leave with a diagnosis and a scope, or an honest no.