I lead agentic software development at BMW Group, building the agent tooling BMW's software engineers use in daily work. I stay hands-on where it matters, for example designing new benchmarks.

Before this I built up and led the LLM Lab at BMW Group Research. We took the LLM voice version of the BMW Intelligent Personal Assistant from first prototype in 2023 to large-scale rollout in more than 100,000 vehicles, with a strong focus on low latency and safety research. I also built a research platform that makes LLM and UX experiments fast to deploy and experience in the car, and that the lab and its partners still train and evaluate on. CAR-bench, our benchmark for agent consistency under uncertain input, received an Outstanding Paper Award at ACL 2026 and took 1st place at Berkeley's AgentX competition from Google Deepmind against more than 1,300 teams.

Previously, I co-founded recoro.io and developed a cloud-first, conversational AI product in my role as CTO. 

I did my PhD in multimodal deep learning with Prof. Schuller and still publish, recently mostly on agent evaluation, environments for fine-tuning and GRPO, and safety.

Outside of AI, I am building furniture and electronic gadgets for my flat. I am always happy to hear from like-minded people, research, and projects! Feel free to drop a line at anytime: stappen_[at]_ieee.org*