Developer and architect from Bielsko-Biała. Software since the demoscene days, AI in real work since 2024.
I got my first computer in 1988 – an Atari 130XE and a copy of “Bajtek” magazine. Then an Amiga 500, my first games written at an age when other kids were discovering football. Later the demoscene, experiments with raymarching, OpenGL shaders, demos at competitions. In 2002 I took my first freelance gigs (PHP, SEO), and in 2007 I went full-time in the industry: Java, microservices, telecom, e-commerce, systems architecture. Since 2024 AI has become part of my real workflow – first through Stable Diffusion and prompt engineering, then through my own AI dev tools.
What I do
Architecture and product decisions. AI today writes code faster than we type it with our fingers. An architect’s strong position in 2026 isn’t coding speed – it’s the soundness of the decisions: what’s worth building, in what order, and when to pull back. Someone has to make that call the human way.
Plan-B architect – for people and for agents. Got a dev team? Great. I’m the one your lead dev calls when the direction needs rethinking, when an architecture review is due, or when you simply need a backup for your CTO. With AI in the workflow the role expands – I’m a lead not just of a team of people, but of a team of agents too. Who does what, in what order, how to monitor the quality of people’s and agents’ work at the same time, where to put the safety switch. I don’t compete with your in-house team – I complement it.
Building in public – Refio as proof. I’m building Refio – a local AI agent for IntelliJ. Every architectural decision, every benchmark, every change – documented in public. A product and an open lab at the same time. It’s a continuation of the same line – back at JDD 2018 I gave a talk on DL4J and neural networks on the JVM stack.
Business ↔ developer translator. Most projects fall apart not over technology, but because business and developers speak different languages. “Let’s implement AI” has to be turned into specifics: where, how, for whom, will it pay off, what to do when it doesn’t. I translate both ways.
How I work – with people and with agents
A collaboration usually runs through four stages:
- Analysis and strategy. I start by understanding the business problem, not from a list of technologies. Sometimes it turns out that “let’s implement AI” isn’t the right solution – and that has to be said too.
- Design and MVP. Fast verification of the key assumptions. With AI agents in the workflow I can run a proof-of-concept in days, not weeks. Where AI actually helps – I use it. Where it doesn’t – I use classic methods.
- Development and testing. Iteratively, with architecture and code review, with monitoring of the agents’ decisions. A human factor at every critical step – without it AI generates code you then have to rewrite.
- Deployment and support. Optimization, monitoring of quality and cost (especially with local LLM models), long-term architectural support.
Technologies and skills
In brief, for those who prefer a table over a paragraph:
Main stack: Java, Kotlin, Python, TypeScript. Spring Boot, React, FastAPI. Microservices, DDD, hexagonal architecture.
AI and LLM: local models (Ollama, LM Studio), Qwen, DeepSeek, local RAG, prompt engineering, fine-tuning on my own hardware. Native function calling, MCP, agentic workflows.
Databases: PostgreSQL, MySQL, MongoDB, Redis.
Frontend / UX: React, Ant Design.
Infrastructure: Google Cloud, CI/CD (Jenkins, GitHub Actions), Gradle, automation (n8n, Make).
On the side: PHP, C++, Bash, x86 Asm, HTML/CSS, OpenGL/shaders.
Who I’ve worked with
Experience with Ailleron, Connectis, Expandi and a few other organizations, among others – telecom, e-commerce, dev tools, automation.
Contact
Got an idea for a project? Need an AI architecture for your team, or an audit of an existing AI workflow? Get in touch – first we’ll talk about whether it makes sense, and only then about the details.