I build AI agents that do real work, and the tooling that makes them reliable. Previously in AI & IoT consulting at Concept Reply in Munich, where I built an internal AI onboarding agent and co-created Skill++. Open to internships in industrial AI from November 2026.
My way into AI came through engineering: a B.Sc. in Industrial Engineering and Management from TU Berlin, and a bachelor's thesis on predictive maintenance for autonomous delivery robots.
- Skill++, co-creator. It watches Claude Code sessions on your own machine and turns the procedures you repeat into reviewed agent skills. Watching and matching run on local models, so they cost no tokens; measured on 30 public recordings.
- Agentic onboarding assistant (internal, Concept Reply). Google ADK and CopilotKit over AG-UI, live company knowledge through MCP, and an evaluation harness that combines contract tests with live behavioral scoring.
- Three-agent proof of concept for an industrial client, covering use cases across the product lifecycle; approved for implementation.
- Teaching: a seven-session course for developers on agentic AI with Google ADK, MCP servers and Ollama.
- Predictive-Maintenance: anomaly detection and remaining-useful-life estimation for autonomous delivery robots (PyTorch), the topic of my bachelor's thesis.
- Agents: Google ADK · Claude Code · MCP · CopilotKit / AG-UI
- Models: Gemini · Claude · local LLMs with Ollama
- Backend: Python · FastAPI · PyTorch · pytest
- Frontend: TypeScript · Next.js · React · Tailwind CSS
German and Turkish (native) · English (fluent) · French (working knowledge)