Summary
James Duncan is a climate-focused machine learning researcher and software engineer with 11 years of experience bridging statistical theory, high-performance computing, and production ML systems. Currently a Young Investigator in Climate Modeling at AI2, he applies stable and interpretable ML methods developed during his PhD work at UC Berkeley to improve climate and precipitation forecasting. His background includes state-of-the-art generative modeling for high-resolution global precipitation (NeurIPS workshop), distributed training on NERSC Perlmutter, and building reproducible scientific software like simChef and VeridicalFlow. He combines deep statistical programming (R, C++, auto-diff) with cloud and HPC operational skills (GitHub org management, CI/CD, AWS), making him fluent from research prototypes to scalable deployments. A less obvious strength is his pedagogy and operational leadership—he has led HPC user support, CI/CD-driven documentation redesign, and teaching roles that translate complex methods into usable tools for diverse teams. Based in Berkeley, he focuses on interpretable, stability-aware approaches to climate uncertainty and model reliability.
11 years of coding experience
9 years of employment as a software developer
Doctor of Philosophy - PhD, Biostatistics, Doctor of Philosophy - PhD, Biostatistics at University of California, Berkeley
Bachelor of Arts - BA, HISTORY, Bachelor of Arts - BA, HISTORY at University of Wisconsin-Madison
Non-degree post-baccalaureate studies, Mathematics, Statistics, and Computer Science, 4.0/4.0, Non-degree post-baccalaureate studies, Mathematics, Statistics, and Computer Science, 4.0/4.0 at University of Illinois at Chicago
Spanish, Japanese