About

I'm Hendrik, an engineer based in Dresden. I build applied AI systems and I care about the part that usually gets skipped: making them scoped, auditable, and safe to hand real permissions to.

That interest isn't academic. I spent 2026 at the European Commission's DG DIGIT, in the unit responsible for data, AI and innovation, working on EU AI Act compliance, governance frameworks for agentic systems, and guidance for public-facing AI assistants. Most of the engineering conversation around agents is about capability. The harder problem is what happens when an agent is wrong: who authorised the action, what it touched, and whether anyone can prove it afterwards.

Before that: production Go and Kubernetes at SAP, product and automation work at Carelytix, and a research assistantship at TU Dresden's chair for immersive media, plus an exchange semester at UiT in Narvik. I'm currently finishing an MSc in Computer Science at TU Dresden. My thesis is a real-time voice agent, streaming speech-to-text, retrieval, an LLM, and speech synthesis over WebRTC, in German and English, where the interesting constraints turn out to be latency budgets and interaction design rather than model choice.

Outside of coursework I run a small homelab: a dashboard written in Go on a Raspberry Pi, a GPU node doing local inference, and an agent layer sitting on top of both. It's a testbed with a purpose. Every claim I make about approval gates, tool scoping, and agent memory is one I've had to implement somewhere first.

After my degree, I am looking for an applied or agentic AI engineering role. If you're building AI systems that will eventually have to answer to a regulator, that's the conversation I want to be in.

This site is fully owned by me and self-hosted: a Go binary with the built site embedded in it, behind Caddy, running on a Raspberry Pi. No analytics, no cookies, nothing loaded from a third party. If you want to talk about a project, or you think I got something here wrong, send me an email.