At the State Audit Office of Hungary, I built a 500-question gold set with HR to evaluate our knowledge-graph RAG, then added automated tests and a composite score for management decisions. The system is used daily by about 200 people, and its agentic retrieval checks whether the available context is sufficient.
I work between domain experts and technical teams, turning requirements into systems people can use. I built the LLM server backend and MCP tool server, and I lead AI delivery in practice by setting technical direction and coordinating work with stakeholders.

