Mehrdad Salimitari

Bio:

I am an AI/ML research scientist and engineer working across computer vision, vision-language and generative models, and the production engineering that puts them somewhere real. My background spans medical AI, autonomous driving, energy systems and production software engineering.

I am currently a Postdoctoral Fellow at the University of Maryland School of Medicine, in the Trauma Radiology Artificial Intelligence Lab (TRAIL). I built a patent-pending system that quantifies traumatic hemorrhage burden and converts it into a clinical urgency score for intervention, now deployed at the R Adams Cowley Shock Trauma Center, where trauma radiologists review and evaluate its output. I own every stage behind it: the labeled dataset built from scratch and linked to a clinical registry of interventions, transfusions and outcomes; the 2D and 3D segmentation and classification models; the per-case uncertainty quantification that makes the output explainable rather than a bare score; and the evaluation gate that decides what ships.

I built the first 3D vision-language model for whole-body trauma CT, on a Stanford Merlin backbone, covering 83 traumatic injuries with no manual annotation (SIIM-CAIMI 2026). Its supervision comes from routine radiology reports, through a production LLM pipeline I designed that runs on new cases as they arrive and keeps extending the dataset on its own. I also co-developed SALIENT, a wavelet-domain diffusion model that generates paired image and mask volumes, so findings too rare to collect still get the training data they need (arXiv, 2026). Both are the same capability: when the labels do not exist, I build the system that creates them, then the models that learn from what it produces.

Previously, as a Postdoctoral Researcher at the University of Denver, I designed multimodal deep learning models for grid-induced wildfire risk prediction and virtual power plants, fusing satellite imagery, weather and grid data. Before that I led backend engineering and security for two consumer fintech platforms, directing 8 engineers within a 23-engineer team of in-house, offshore and external collaborators.

I received my Ph.D. in Computer Science from the University of Central Florida under Prof. Yaser Fallah, working on cooperative perception for connected and autonomous vehicles. I built the multi-vehicle sensor platform behind it, solved the time synchronization and spatial registration that made its data usable, and showed that transmitting compact learned features instead of raw point clouds cuts V2X bandwidth while improving detection accuracy. During my studies I also collaborated with the Center for Research in Computer Vision (CRCV) at UCF under Prof. Nazanin Rahnavard, developing sample and feature selection algorithms for efficient deep learning. My research has been supported by NSF, Toyota and Cyber Florida.

What interests me most is the gap between research models that score well on benchmarks and systems that actually hold up in production.