At Sportradar, I designed the evaluation framework and acceptance criteria used across five concurrent AI training programmes. I also deployed ML and LLM automation that replaced 70% of manual review, measured against a defined quality bar.
I built InVerba as its sole architect and engineer: a multi-tenant RAG document platform with grounded, cited answers and per-tenant security controls. I also built open-source projects for document Q&A with groundedness scoring and cost tracking, and a multi-agent research workflow with a reviewer quality gate.

