Examples

A few proofs.

Three AI implementations I worked on from the inside — what they were about, what actually shipped, and what they mean for your process.

Morizon/Gratka: natural-language search

co-founder & CPO, Morizon-Gratka Group

Plain language as a better interface to complex search.

Real estate search engines are forms with a dozen filters — but people think in sentences: “looking for a studio to rent in Kraków under 3k”. In May 2023 we launched mieszkAI on Morizon.pl and szukAI on Gratka.pl — the first natural-language property search on the Polish market, built on GPT, with voice search on mobile.

What shipped

mieszkAI (Morizon.pl) and szukAI (Gratka.pl) — the first deployment of its kind in Poland
query interpretation mapped onto listing structure: location, price, area, amenities
voice search on mobile devices
first test version in a few weeks; the real work was weeks of analyzing actual user queries

The process

plain-language query → interpretation → mapping to data → results & recommendations

Scale & effect

220,000+ listings covered by the search
10,000+ AI-powered searches in the first month
AI-search users returned more often and sent more inquiries to listers

TasteRay: agents in the team’s daily work

co-founder & CTO

The agent executes the ticket; the human owns the quality.

TasteRay is AI-based movie and TV recommendations — it matches titles to who you are, not what’s trending (iOS, Android, web). On the engineering side it’s the proving ground for the working model I sell in sprints: coding agents are a normal part of the team’s daily process, with explicit autonomy boundaries.

What shipped

product: recommendations with an explanation of why a title fits — on iOS, Android and web
process: tickets written for agents, repo instructions, AI + human code review
autonomy boundaries in practice: the agent proposes and builds, the human approves and owns quality

The process

ticket → coding agent → AI review → human approval → shipped

Scale & effect

the working model runs in production, in the team’s daily rhythm
the same process is what I transfer to companies in the acceleration offer

Wondel.ai: AI for everyday people

founder

The hardest part of AI products isn’t the model API — it’s UX, pricing, distribution and trust.

Wondel.ai is a Warsaw-based studio of consumer AI products: text, image, data and developer tools. The best-known product — czat.ai — is conversations with specialist characters (from a dietitian to an IT expert), no account, nothing to install. Simple, paid, useful.

What shipped

czat.ai: dozens of specialist characters in one simple interface
a product family: Czat.ai, Obrazki.ai, Firmy.ai, DeveloperToolkit.ai
one UX across many models (GPT, Claude, Gemini) — the model is swappable, the experience constant

The process

idea → paid product from day one → iteration on real usage → scale or kill

Scale & effect

200,000+ czat.ai users
millions of conversations
a product portfolio built and run by a small team

Want results like these?

Book a call — 15 min Check if your process is Sprint-ready first