Senior full-stack engineer · Systems that hold money and time · Remote
Backend engineering since 2012. Most of it agency work on e-commerce platforms — high-traffic stores processing transactions and customer data, including Dyson’s. The kind of systems where an outage means lost orders rather than a failed build. Four of those years were on day rate, including a booking-and-payments platform I designed, built and then ran for the same client for four years.
Then I stopped. Six years travelling, March 2020 to March 2026, no software in any of them. What brought me back was coding agents — I could suddenly build the things I had only ever been able to spec. What I build and operate now is a live scheduling, payments and video platform — web and Android, Next.js, React Native, Postgres, Stripe Connect with cross-border payouts — carrying real bookings and real money.
I bring the delivery pipeline with me: a merge gate that has to be green before anything lands, a coverage floor that only ratchets upward, migration-drift and deployment checks, and the decision records that say why each of them exists. The two systems written up below are described with their trade-offs and mistakes intact, not as a portfolio list.
And I build almost entirely with coding agents — more than 8 in 10 commits across those systems were written or co-written by an agent, counted from git history, not estimated. The same sessions provision the servers, run the deploys and maintain the decision records the next session reads. It is mostly a story about verification rather than generation: what I built is the system they operate inside. How that works is stated plainly rather than left to be inferred.
The evidence
Everything this site argues rests on artifacts you can inspect: figures counted from git history by script, a practice written up with its failures included, and a tool you can run yourself. Where a number is an estimate, it says so.
One operator’s systems are a starting point, not an answer. The point of measuring mine honestly is to have something worth comparing with yours.
The agent-operated practice this work runs on, and the two products it ships — one in production carrying real payments, one built and parked pre-launch. All described with their limitations intact.
The practice both systems here are built with: parallel agent sessions, gates that do not depend on reviewer attention, and the context layer that turned out to be lying.
Read the case study
Scheduling, payments and live video teaching for independent language teachers, live in production since May 2026.
Read the case study
Inferring where buses are from anonymous rider signals, on a network with no vehicle tracking, no published timetable and no operator data feed.
Read the case study
Shorter pieces on specific problems.
Senior TypeScript, Next.js and Postgres — most useful where the hard part is the system rather than the screen, and where I would be reviewing other people’s code as well as writing my own. I contracted for four years before the break, on two concurrent engagements, so this is a return rather than a first attempt.
UK-based and available now. Currently on a temporary working trip abroad, with a daily overlap from about 13:00 UK and flexibility outside it. Back in the UK early in the engagement, and on site from then.
And if your team builds with agents, I want to hear how it is really going — twenty minutes, nothing sold, notes compared both ways. That conversation starts here.