Sixteen years, 500+ projects, and one recurring pattern: the hard part is almost never the code. It's the gate someone else controls.
A health authority deciding whether a telehealth platform may operate. A payment certification body deciding whether a card reader may take a transaction. A procurement committee deciding whether a 1,400-camera architecture is affordable over three years. A field engineer at 2am with no signal, deciding whether the app in their hand is worth trusting.
None of those are engineering problems in the usual sense. All of them decide whether the engineering was worth doing.
That's the work I've ended up specialising in — not a language or a stack, but products that have to satisfy someone outside the building. It turns out the discipline transfers cleanly. A panic alarm that must fire without a network and a foreign-object detector that must not miss a hand are the same problem wearing different clothes: be right under conditions you don't control, or the thing you built is worse than nothing.
The bench is easy. The bench is always easy.
On a recent computer-vision job I identified the camera, the runtime and the loop rate from screenshots alone, before reading a line of source. Constraints found early are the ones that stop you proposing a fix the hardware can't run.
When a detector both misses real objects and fires on nothing, lowering the threshold is not a fix — it's choosing which way to fail. Symptoms point at causes; they rarely name them.
Milestone-paid, delivered against something we both signed. There's a delivery method behind it from nine years of agency work, but the method isn't the product. The shipped thing is.
A system that needs me forever is a system I built badly. The telehealth platform has run for years on the client's own team with no dependency on me. That it survived without me is the proof.
Not scarcity theatre — that's genuinely the number at which I can stay hands-on rather than becoming a person who forwards emails between other people.
I use AI-assisted tooling heavily and it compresses build time considerably. What you're paying for is architecture and judgment. Pricing on hours would be pricing the wrong thing.
The detail sits in the project archive — fifteen systems, each with the gate it had to clear.
Worth stating plainly, because it's the part most engineering CVs can't claim:
Clients are unnamed by default. References and live links on request.
I'll tell you in one paragraph whether the problem is where you think it is, and whether a scoping week is worth your money. No charge for that, and no pitch if the answer is no.
Start there →