Amstelveen, Netherlands · Independent, via ZZP
Principal Software Engineer / AI Researcher
I take systems all the way to production — the platform underneath, the infrastructure around it, and the product itself. Thirty years of doing that, most of it on the parts other people would rather not own.
These days that work runs on AI leverage: agents handle the breadth, small teams keep the judgement. Rarely alone, and never unverified.
Infrastructure as code, cloud and hybrid, CI/CD that people actually trust. Multi-region and self-healing when the workload earns it, boring and cheap when it doesn't. Monitoring designed so that silence means healthy, not dead.
Building with AI agents as part of the delivery process rather than as an autocomplete: specification first, agents for mechanical breadth, and a verifiable invariant behind every claim that something is done.
Backends, databases and the unglamorous integrations that decide whether a product is real: payment providers, fiscal and tax compliance, point-of-sale and booking systems, third-party APIs that were never meant to be automated.
The physical half that cloud engineers usually hand off: structured cabling, VLAN segmentation, Wi-Fi design, captive portals, site-to-site links. Useful precisely because it sits on the same side of the fence as everything above.
"AI-driven" is doing a lot of unearned work in most job adverts, so here is what it means concretely in mine.
An agent will happily produce a confident answer to a question nobody checked. So the spec comes first, in writing, and the work is measured against it — not against whether the output looked plausible.
Exploration across a large codebase, migrations, test scaffolding, review sweeps: this is breadth work, and it parallelises well. Architecture, trade-offs and knowing which risk is acceptable do not. I don't pretend otherwise.
"Done" means an invariant was defined and then confirmed — a test that could have failed, a query that could have returned the wrong number, a health signal that can tell idle apart from broken. Anything less is a claim, not a result.
The interesting systems have owners, operators, accountants and regulators attached to them. Most of my work is inside small teams and alongside the people who will run the thing after I leave — the leverage is in the method, not in heroics.
A restaurant, hotel and event venue running on one system I designed and built out: QR ordering and bill payment for guests, payments split across separate legal entities, fiscal receipts through the relevant tax devices, live integration with the existing point-of-sale, room and pergola bookings synced to OTA channels, and a Telegram assistant for guests on top.
Underneath it, the same engagement covered the physical site: network segmentation, Wi-Fi, a captive portal for guests, and VPN routing so payment providers stay reachable.
Transaction indexing and wallet infrastructure on a high-throughput chain: ingesting hundreds of millions of transfers, exposing them through an API with predictable latency, and delivering webhooks reliably enough that downstream systems can settle money against them.
Long-running platform and reliability work inside large organisations — most recently in the energy sector, earlier for a European payments company and a digital consultancy serving millions of requests a day. Immutable infrastructure, hardened images, least-privilege IAM and fully automated multi-repo pipelines.
Based in the Netherlands, working independently via ZZP alongside my current role. Happy to talk through a problem before there is a scope.