From a garage to an AI builder
When I solve architecture, I see the P&L consequences. When I talk to investors, I talk about KV cache too. This combination is rare — and that's exactly where I'm useful.
In 2002 I founded Hosting90 in a garage. For the next eighteen years I built it into a real infrastructure company — domains, hundreds of servers, the whole provisioning and reliability stack — with a team of about twenty-five. I led people, wrote orchestration software, did Linux admin and senior dev work. I sat in the CEO chair, where understanding code and cash flow at once wasn't a choice but a necessity. In 2020 I sold the company to WY Group.
Then I took a year off and went deeper into AI. The longer I worked at it, the more one gap interested me: the chasm between what AI labs publish and what a solo founder can actually afford and run on their own infrastructure. That gap is what I enjoy most.
So I went all in. I'm a core contributor to the open-source inference engine vllm-mlx — I work on the economics of running AI models: when it pays to run locally (private data, production savings on high-volume work) and when to reach for the cloud. I also built my own scraping infrastructure — nine LTE modems and a few Raspberry Pis. I don't do deep tech for show; I do it out of necessity, so things make economic sense.
As CTO I built the technical architecture and AI stack of three production AI products. And earlier I co-founded GuruWatch — used today by brands like Lenovo and Niceboy — and Lobot, which still runs today.
And then came the thing that brought me here. As someone who built, led, and sold a company, I learned the modern AI tools for software development — and saw firsthand how fast you can build with them when you know what and how. I've helped several business owners build their own internal apps. Things they had quotes for in the hundreds of thousands suddenly made sense to build themselves, under my guidance — and to own.
That's my work today. I don't just bring code — code is a commodity now. I bring judgment about what's worth building and what isn't, and the depth that makes the result secure and able to survive real operations, not just a demo. I talk about ROI and goals as naturally as about how to build it technically.
I'm based near Prague and I write about the economics and infrastructure of AI. If you have an idea in your company for a tool that would save you work or money — I'm happy to tell you whether today's AI can build it for you, and to guide you through it.
I wrote about my scraping infrastructure — the nine LTE modems and Raspberry Pis — in the LTE proxy pool article.
