Soroush Dianaty

I trained and practiced as a Family Physician (M.D.)—treating over 7,500 patients across 17 rural and urban communities—before transitioning into computer science and biomedical informatics research.

During my clinical practice, I experienced firsthand how traditional clinical decision support tools fall short when handling real-world ambiguity, non-standardized EHR notes, and dynamic patient trajectories. As Large Language Models began entering healthcare discussions, it became clear that evaluation metrics borrowed from general NLP (like exam accuracy on multiple-choice questions) fail to capture the asymmetric risk of clinical hallucinations.

Now as a Biomedical Informatics PhD Researcher at Arizona State University, my work focuses on building the mathematical, computational, and standards-compliant frameworks necessary to evaluate generative clinical AI before it touches patient care.