Zeeshan Zia

I build AI systems that understand real-world activity. At Retrocausal, I focus on visual activity understanding, enabling models to interpret complex human workflows in industrial environments. Our platform is deployed with leading manufacturers and grounded in direct exposure to real operations.

My research centers on how neural networks represent structured processes over time, including unsupervised alignment, action segmentation, and temporal reasoning. I have authored 50 peer-reviewed papers and hold 25+ patents in activity understanding and representation learning. My lab is recognized for leading work in unsupervised activity segmentation and alignment.

I am broadly interested in how learned representations capture structure in complex systems and how to translate that understanding into scalable, real-world AI products.