Computers have fascinated me since I was a child. It took me a long time to find the courage to make more than a hobby out of it. Perhaps that is exactly what makes the difference.
With a clear plan in mind I started vocational training as a Media Designer in Herzogenaurach (near Erlangen). The idea: learn UX and UI design, from grids and typography through to brand management. What I did not see coming: between the layouts, programming questions were already running alongside me, and I would not be able to dodge them forever.
In the middle of the pandemic I was suddenly expected to build a project I had designed myself, a WordPress template from scratch. HTML, CSS, JavaScript and above all PHP arrived all at once. It felt like swimming in a lead vest. But the deeper I dug, the more I understood that I enjoyed the logic underneath. Duty turned into curiosity.
At 25, with my training just behind me, it was clear to me that this could not be all of it. I left my home town of Erlangen, moved to Dresden and started a dual-study programme in Media Informatics. Only there did the pleasure of building meet the theory behind it, and a lot of what had been gut feeling finally got a name.
Three years later I was holding the B. Eng. in my hand. Since then I have been at Digitalwert in Dresden as Innovation Manager for Artificial Intelligence and Software Engineer. My brief: build AI in where it makes projects measurably better, and at the same time deliver solid software that works without any buzzwords.
My path runs from design through the craft to software and automation engineering. Today I carry both perspectives in the same toolbox: I know how a product should feel for the person in front of it, and I can build it cleanly. Working at exactly that intersection is what I enjoy most.
Three strands of my work, from production pipelines to my own inference environment.
Headless CMS and website content vectorised through automated pipelines and made usable as chat assistants, including an admin UI for synchronisation and RAG parameters.
Contributed to an opt-in based system that evaluates user signals in a structured way and gives teams qualified insights for customer outreach.
A GPU server running self-hosted open-source language models. I use it to evaluate RAG configurations and model behaviour before they make it into projects. Alongside that, I have experimented with Whisper for speech-to-text and researched open-source diffusion models for image generation on my own servers.
N8N is my tool of choice when systems need to talk to each other but nobody wants to write the glue code. From lead enrichment through content pipelines to RAG sync jobs for vector stores. Workflows are versioned, tested and run on my own infrastructure, not in the SaaS cloud.
Automated pipelines that vectorise website and headless CMS content, refresh embeddings and keep the RAG index clean on every change. Including error handling, retry logic and logging.
Contact data and signals from several sources are merged, classified and fed into CRM or marketing tools. LLM-based classification can optionally run on self-hosted models.
Slack and mail bots that pre-qualify tickets, compile status reports or trigger routine quality checks. Small workflows with a lot of leverage.
I am a Full-Stack Developer and Forward Deployed Engineer with an academic background in Media Informatics (B. Eng.). My focus is on building modern software and integrating production-grade AI systems, from the data layer through to a usable interface.
Frontend with Nuxt.js, Next.js and Astro, backend in Laravel and Nest.js. Data lives in Postgres or plain MySQL, infrastructure runs on Docker. Mobile app development with React Native and Flutter, desktop development with Tauri and Electron.
Short release cycles, clean version control, documented interfaces. I work closely with designers and stakeholders and propose the simplest setup that carries the requirement (MVP). No frameworks for their own sake, no microservices without a reason, no AI systems where a spreadsheet would do.
Analytical thinking, structured problem solving and strong communication skills. I translate between development, design and stakeholders, and turn complex requirements into technical solutions that hold up.
AI is a tool, not a substitute for judgement. I use modern AI-assisted development environments such as Claude Code, Codex and Cursor productively and as a matter of routine, but I insist on a clear separation between automating the craft work and the domain decisions, which in my view should always stay with people. Current focus areas: generative frontends and the responsible use of language models and diffusion models in production applications.