AI that keeps working after the demo.
I'm Ros. I spent about 20 years building software teams, the last few of them as a CTO. Now I build automation and data systems on my own, mostly for businesses where one process quietly eats a day of someone's week.
What I take on
Two kinds of work.
01
A week finding out whether AI helps you at all
I look at how the work actually runs, talk to the people doing it, then write it up. What's worth automating, what it costs to run afterwards, and what I'd start with. Sometimes the answer is that it isn't worth doing, and I'll tell you that.
02
One job taken off someone's desk
An automation that handles the whole job, start to finish. It runs on your systems and it tells you when something goes wrong. Documented well enough that you don't need me to keep it running.
Background
How I got here.
The last few of my twenty years in engineering were as CTO of a payments company, with around 60 engineers. The platform went from 20,000 to 150,000 requests a minute while we moved it off a monolith.
I left in the middle of 2025. The honest reason is that I'd stopped building. My days were meetings, hiring and escalations, and I'd become someone who talks about software.
There was a less noble reason too. I could see what AI coding tools were turning into, and I wanted to find out what one person could do with them. That's hard to test in a job where your value is measured in abstract numbers with no connection to the result.
Something I built
Risheno
Risheno lists solar and battery installers in Ukraine. Picking the wrong one costs thousands, and every company's website says roughly the same thing about quality and experience.
So every company gets checked against the state business registry before it goes live. Some stay marked as unverified, and I show that rather than letting people assume everything has been checked.
I built and launched it on my own, directing several AI coding agents. That only worked because of the guardrails I put around them. Left alone they make quiet mistakes and then report success.
Who I work with
Two kinds of client, and they arrive with opposite problems.
Companies where nobody has automated anything yet. Often not software businesses at all, and that's usually where the biggest wins are sitting, because nobody has looked.
Also engineering teams already using AI coding tools. More code than ever, and less certainty about it. I've had that problem myself, and I built the checks I wanted before I had to sell them to anyone.
Fixed scope, fixed price, and dates I keep to. Remote, in English or Ukrainian.
Got a process worth automating?
Or data that's turned into a mess. Fifteen minutes on a call is usually enough for me to tell you whether I can help.