At BMW, I’m building ChatBMW’s persistent car memory, including structured cross-session recall, driver isolation, audit logs, and a dashboard. I also built the Welcome Agent end-to-end and integrated a screen-reading GUI agent through MCP adapters, PubSub, and the OpenAI Realtime API.
I design multi-tool, function-calling agents with RAG and agent-to-agent orchestration on AWS Bedrock, deployed across AWS and Azure on Kubernetes. My production work includes container-based CI, sandboxed tools, allowlisted commands, and path validation.
Previously, I improved data workflow latency by 70% at American Express and built clinical data pipelines and dashboards at Mango Sciences. My research at Forschungszentrum Jülich focused on generative models for inverse design, with PyTorch and TensorFlow pipelines for GANs, diffusion models, evaluation, and scientific experimentation.
