Josip Vrdoljak, MD, PhD
Clinical Research
Josip holds an MD, a PhD in evidence-based medicine with a focus on medical machine learning, and an MSc in Artificial Intelligence — a combination that is genuinely rare: clinical judgment grounded in practice, research depth grounded in published evidence, and engineering capability grounded in production deployment.
His research at the University of Split School of Medicine focuses on evaluating and improving state-of-the-art large language models for clinical decision support and medical error detection — work that requires understanding both what a model produces and what a clinician actually needs.
On the engineering side, he has delivered production LLM solutions as a lead ML and backend engineer: applications built on RAG architectures, dynamic tool and function calling, and voice-to-voice agent systems deployed on AWS and GCP. Most recently, he led the end-to-end fine-tuning of MedGemma 27B for medical error checking — combining supervised fine-tuning on synthetic data with RLVR (reinforcement learning from verifiable rewards) using clinician-curated datasets, targeting deployment in a live hospital environment.
Within the collective, Josip works at the intersection of clinical expertise and production AI engineering — the place where healthcare organizations need the most help: not just AI that sounds correct, but AI that can be trusted, evaluated, and safely deployed.
MD · PhD, Evidence-Based Medicine (Medical Machine Learning) · MSc, Artificial Intelligence · Postdoctoral Researcher, University of Split School of Medicine