Mladen Turkovic

Resolve a production quality failure: strong internal metrics, consistent end-user complaints. - Conducted end-to-end analysis of workflows, and failure cases to identify root causes - Replaced aggregate accuracy reporting with a 5-number summary to expose worst-case failures; implemented via Python/MLflow - Built a JSON comparison engine for structured LLM output validation and regression testing - 5–25% improvement in median accuracy/recall; up to 30% error reduction in critical extraction fields

end-to-end NLU/NLP engineering across 12–18 languages and 5 vehicle brands (VW, Škoda, Seat, Audi, Porsche) - Production system deployed across Audi and Porsche vehicles (SOP 2020–2025) - Designed and maintained reproducible evaluation pipelines for iterative NLU model improvement - Directed development of internal automation tooling (Python, PowerShell, Bash, R) covering NLU - Owned multilingual data collection, quality governance, and dataset restoration across 11 languages