I’m Janusch Koza, software architect with 30 years of engineering experience. I work across domains — industrial automation, logistics, publishing, healthcare — wherever complex technical systems need to be understood quickly and built to last.
Focus
- System diagnosis and architectural decision-making
- Modernising production systems (.NET, WPF, WinForms, ASP.NET Core, distributed)
- Memory and performance analysis of production applications
- AI integration into existing production systems
Capabilities
What runs through all of my work: taking tangled problems apart until the actual root cause is visible. I don’t treat symptoms. I trace the connections behind them — with measurement and traceable reasoning.
The .NET ecosystem is my primary tool — from desktop applications and backend services to runtime and performance analysis. Architecture and clean design come before any single technology: sound structures, clear interfaces, and a reliable path from grown systems toward modern solutions.
On the data side I work with relational, object, and vector databases. I integrate AI models into production applications — RAG pipelines for similarity search, but increasingly also knowledge graphs with typed relations, where an AI assistant needs to reason over a domain instead of just retrieving text that resembles it. That distinction, retrieval versus reasoning, is what decides whether an AI-assisted workflow holds up in production or only in a demo.
I have a particular fondness for well-structured DevOps: CI/CD pipelines, reproducible deployments, and automated workflows that make the path from development to operations dependable.
Kybron
Alongside consulting I’m building Kybron, an independent product under the Lab4SR name. It starts from a conviction thirty years of building systems have given me: the structure of processes repeats, their content doesn’t. So Kybron’s kernel deliberately knows nothing about what it runs. What a step actually does is decided by small, replaceable building blocks, each made for exactly one case instead of a universal solution configured to fit. AI helps build those blocks; running the plan stays rule-based, traceable, and auditable. Kybron isn’t finished, so I show it in stages — the first chapter is on the blog.
Industries / Domains
- Industrial and automation technology
- Mechanical engineering and plant construction
- Logistics
- Healthcare
- Publishing & media
Languages
- German — native
- Polish — native
- English — fluent
Away from work
When the computer is off, my full attention belongs to my wife and my grandson.