Computational data analysis

Modern astronomy is, increasingly, a data analysis problem. Surveys now deliver spectra and photometry for hundreds of millions of stars, and the bottleneck is rarely the telescope — it's our ability to model that data faithfully, at scale, with honest uncertainties.

A lot of my research is methodological: building models and inference machinery that are fast enough to run on enormous datasets, but principled enough that we can trust what comes out. That spans data-driven models (learning the mapping from data to physical labels), careful likelihood-based inference, and the unglamorous-but-essential work of making methods numerically stable and reproducible.

Themes I keep coming back to:

• Scalability without compromise — methods that work on millions of objects but still respect the physics and the noise model.
• Honest uncertainties — propagating what we don't know, not just reporting what we do.
• Open, reusable software — most of these ideas only matter if other people can pick them up and use them.

Because these tools are general, this work has let me contribute across a surprisingly wide range of domains — from stellar spectroscopy to gravitational waves.