I'm not a consultant who hands over a document and leaves — when it makes sense, I build the solution too.
AI consulting → Projects cz-agents.dev →For twenty years I've built software where mistakes get paid for in certification and audit — payments, banking, industry. I now build AI integrations for clients in the US and the UK as well. I apply that same discipline to AI: my own systems over regulated public data, running in production, not in a demo.
I help mid-size and larger companies decide whether and how to adopt AI — before it costs millions. And when the decision makes sense, I take it into production.
AI over regulated public data, built for production, not for demos.
Czech company and court registers turned into data you can actually point AI at — with nothing left for the model to invent. Nine servers in production, open source, with a paid layer on top. This is exactly the class of problem you face the day you point a model at your own internal systems.
European grant programmes turned into a basis for deciding who to partner with and what to go after — 27,000 projects, 16,000 organizations, six languages. Running in production, connected to official EU sources. Second domain, same method: „AI over regulated public data“ isn't a one-off here, and on the OP TAK project I wrote the proposal and the business goals.
Hunting distressed real estate — daily crawl of the insolvency register and auctions, Vision AI extracts data from PDFs, scoring ranks leads by real value. Proof of an AI pipeline over messy public data.
Electronic mileage logbook for Czech companies and freelancers. AI receipt recognition, statutory reimbursement calculation, export to Pohoda, Money and iDoklad. Proof of working with legislation and Czech accounting integrations.
Alongside client work I build and run community apps — a civil-defense shelter map, a wheelchair-accessibility map and more. → mhai.app
Not because AI couldn't do it. Because that particular thing has a better and cheaper answer elsewhere — connecting two systems that don't talk to each other, one export, or simply not retyping something by hand any more. What's left is one or two ideas worth doing. And those are worth doing properly.
I ask about it in the first meeting: what has to happen within three months for you to say this was worth the money? When nobody in the company has an answer, the problem isn't AI. It's a sign not to start yet.
A chatbot on the website looks good in a board meeting. But the hours almost always disappear into work nobody puts on a slide — retyping data between systems, digging through email, checking the same documents by hand over and over. Dull things. And that is exactly where the hours come back.
Not if, but when. I spent twenty years building systems in banking and industry where a mistake is paid for in an audit — and it taught me that the difference between a demo and daily operation isn't how it looks, it's what happens when something doesn't add up. Cheap quotes leave that part out, because nobody can see it. Until an inspection shows up.
Writing the app is the easy part. Hosting, security, backups, GDPR, payment gateway, ongoing maintenance — that's the part most prototypes don't survive.
Tell me what's giving you trouble. Within two days I'll tell you whether AI solves it, or whether you should keep your money. That answer is free.
[email protected]