The engineer who also builds.

Who I am

I have been independent for over 20 years. I started as a young freelancer in Vienna; today I run my own company and a network of experienced developers. Along the way I worked on more than 100 projects and wrote, reviewed and shipped code with over 300 developers.

Vienna is my base, the work is remote: for clients across the DACH region and beyond. My thinking stays Viennese. Direct, a little self-deprecating, with a healthy distrust of anything that sounds too smooth.

My approach

I am the architect who also builds. When a client says "we need feature X", I first ask: does X actually solve the problem? Only when the requirement is right do I write code.

On one project the team gave me a nickname: Worry Eater. I grab the problems nobody wants. Performance bugs, legacy code, edge cases in production. Not out of masochism. That is where the biggest leverage sits.

The result: queries down from minutes to under a second. Not with more hardware, but by understanding what the database actually does.

Don't guess, measure

In AI projects the demo does not decide anything. What counts is the behavior after the third prompt change. That is why I build evaluation in from day one.

On a RAG project my assumption was: vector embeddings are the standard. I built both variants and compared them. LLM-based metadata analysis came out 29% better. And GPT-4 turned out the better choice over GPT-5: lower cost, higher accuracy.

I don't know this from a blog post. I measured it.

AI agents and Pinchy

I don't just use AI agents in client projects, I work with them daily. OpenClaw is my tool for that: autonomous agents that work independently and deliver results.

Out of that experience I am building Pinchy, an enterprise layer on top of OpenClaw. With permissions, an audit trail and a plugin architecture. I know what happens when an agent runs without a control layer. The story is on the Pinchy page. Pinchy is open source and runs in production at Helmcraft.

How I work

I handle most projects on my own. AI does the grunt work, my experience does the rest. That is how I deliver what used to take a team.

I work at a fixed price. You know the price before we start. No add-ons, no nasty surprises.

When a project grows, I bring in professionals from my network. I still work with the best people from 100+ projects, for example at Mit Freuden: four developers with 15 to 20 years of experience each, one direct line. No account managers, no telephone game.

I have plenty of leadership experience, including four years as a CTO. But permanent management roles without code are not for me. I deliver through results, not through meetings.

Expertise

Product development

• Thinking end to end: requirements, architecture, UX, business impact
• Direct communication instead of ticket ping-pong
• Technical decisions made with the business impact in mind

AI and agents

• LLM integration with measurable quality
• Evaluation from the start, not as an afterthought
• Multi-provider architectures: OpenAI, Anthropic, Google
• Agent-native systems: software built for agents

Full-stack engineering

• Production systems with billions of records
• Performance tuning: from minutes to under a second
• Strict TDD, hundreds of tests as a safety net