Mehran Aghabozorgi

I build learning systems that do more with less: generative models that learn well from limited data, sample-efficient reinforcement learning with world models, and fast, RL-fine-tuned vision-language-action (VLA) policies for robots. A recurring theme in my work is multimodal, sample-efficient generative modeling and using it to learn reliably from little experience.

I am a PhD candidate at Simon Fraser University (APEX Lab), advised by Ke Li, and affiliated with Amii. I expect to graduate in Fall 2026. Currently I am a Research Intern at RBC Borealis, where I work on improving the inference speed of multi-billion-parameter multimodal models for robotics. In parallel, I am developing a novel, sample-efficient approach to reinforcement learning fine-tuning of multi-billion-parameter vision-language-action (VLA) models. Before my PhD I spent three years shipping production software, including a verifiable voting system and end-to-end encrypted cloud storage.

I am on the job market for 2026 and always happy to chat about RL, generative modeling, or robot learning.