boloses1 User

At i22 Digitalagentur, I build agentic AI systems for enterprise clients, including a multi-stage content platform with human-in-the-loop quality assurance that reduced API costs and token usage by 90%.

I've redesigned slow monolithic AI assistants into orchestrated systems that dynamically load domain expertise, documented architecture decisions in ADRs, and trained client teams to scale additional agents. I also created reusable MCP microservices for tool calling, CI/CD, containerization, and Traefik routing, now used across multiple client projects.

Previously at Universitätsklinikum Heidelberg, I led GDPR-compliant clinical-text anonymization for the national GeMTeX project, fine-tuned open-source medical models, and built RAG and evaluation pipelines. My work spans medical NLP, transformer training dynamics, and explainable machine learning.