John Hoffman

I'm a staff data scientist at Meta's Fundamental AI Research lab, where I drive evaluation and data strategy for large-scale AI systems. I co-led evaluation for Movie Gen and was a core contributor to SAM Audio, leading all human evaluations. Two of the projects I've worked on — SeamlessM4T and NLLB — were published in Nature. I've also worked on understanding and improving the reliability of our large-scale GPU compute infrastructure.

Before FAIR, I earned a PhD in astrophysics from Princeton, where I developed GPU-accelerated methods for astronomical time-series analysis. My thesis library, cuvarbase, was adopted by NASA's TESS pipeline and enabled a Nature-published discovery of ultracompact binary stars. I also worked in ML consulting and ad-tech before joining Meta.