// about
I'm a software engineer in London. For about ten years I've been the person you hand a system that has to work — a courier network, a trading account with unforgiving rules, a threat-intelligence pipeline — and I stick around until it does. Somewhere along the way I stopped trusting anything that can't prove what it did, including my own code, and started building tools around that suspicion.
Told as eras rather than job titles. Employers are described by what they do, not who they are — partly discretion, partly because “the parcel years” is a better story.
A geospatial software company, twice — I left, had an adventure, and came back. We built transport-accessibility tools for councils and planners: the software that answers “how long does the bus actually take?” I learned multi-tenancy, single sign-on, test-first delivery, and that performance work matters most when a planner is watching a spinner instead of going home.
A courier-management SaaS — the adventure in the middle. Real-time logistics with vans on the road and money on the line. It's where I learned that a distributed system isn't an architecture diagram; it's what happens when a driver's phone loses signal halfway through a delivery confirmation.
An algorithmic trading firm: platform microservices, backtesting frameworks, ML prototypes in Python. This is where the itch behind Shamshir started — the awkward gap between a strategy that backtests beautifully and code you'd trust with actual money. They are, it turns out, rarely the same code.
A data-platform consultancy on GCP: high-throughput APIs, creaking batch jobs rebuilt as event-driven flows, infrastructure as code. Mostly I learned the difference between “it works” and “it works at 3am under load while nobody is watching” — and that most of engineering is quietly closing that gap.
The formative one. Sole engineer at an early-stage AI cybersecurity startup — which sounds glamorous until you realise the backend, the front-end, the cloud, the releases, and the “why is it down?” messages all route to the same person. Event pipelines swallowing everything from firewall logs to 50GB dumps, a conversational threat-intelligence UI, and a RAG engine wired in end to end. Working next to an LLM every day taught me the thing the lab is now built on: the model is rarely the bottleneck — the context you feed it and the checking you do afterwards are.
Before all that (2011–2015): the apprenticeship years — desktop apps, early single-page web, and the discovery that shipping software people pay for is a different sport from writing code.
Most of my spare cycles go into the lab — an orchestrator that runs coding agents without believing a word they say, a code graph that cites its sources, and a trading engine built as a pure function. Different domains, same suspicion: if a system can't show its work, it's guessing. The writing is where those ideas get argued out properly — that's also where the deep technical stuff lives, if this page felt too gentle.
There's no CV download here — the eras above are the long story, and the writing shows how I think, which is usually the more useful half. If you'd like to talk shop, work together, or point out that one of my “nearly done” projects has been nearly done for quite a while, email works.