Fifteen years, three careers, one thread I only saw looking backwards: I keep turning unclear problems into working software — a business that had to exist before anyone wrote code for it, trade-permit systems, client documents, other people's private thoughts. This page is how that happened.
Founded Addqd, then Pinflorist in Riga; ran business development at ZenMall; art-directed at Trade&Chat. Seven years on the commercial side taught me the thing engineering school can't: what a problem costs before anyone writes code to fix it. When I sit with a government officer or a firm's partner today and listen for the expensive problem — that's this act talking.
Built and rebuilt frontends for crypto and investment products — Nebeus, InvestEngine, Vinza — then a React Native product from scratch at iBride. Finance is a good teacher: correctness isn't a preference, it's the product. Alongside the work, a Computer Science degree at the University of London (2020–2024), because self-taught needed a foundation under it.
Senior engineer at Social Discovery Ventures: modernising a legacy platform serving real traffic — component systems, performance, launching a new product on the core without stopping the old one. Greenfield is easy; keeping a living system alive while you rebuild it is the actual skill. It's also where rescue work stopped scaring me.
Two roles, one shape. At Kyrgyz Single Window, I joined as head of IT, moved into frontend architecture, and now lead the AI work — systems for foreign-trade participants and the agencies that issue their permits, built on state infrastructure. As founder of Open Cradle, I'm building an on-premise agent platform: law firms, clinics, family offices — teams whose documents legally cannot touch a cloud API.
Both started the same way: a specific, badly-defined problem, turned into a working system. The data-can't-leave constraint happened to be part of both — it's the proof this approach holds up under real limits, not the whole of what I do.
Live products and experiments — on the projects page.
Most failed AI projects chose the technology first. I interview the people who do the work before I open an editor — sometimes the answer is a form redesign, not a model.
In demos, iterate loud and quick. In regulated systems, deterministic rules decide and every claim cites a source. Knowing which mode you're in is the skill.
A prototype your stakeholders actually use teaches more than a quarter of planning. Real usage data settles arguments that opinions can't.
If software won't fix the problem, that's what the recommendation says, with the reasoning.