These days, I spend most of my time on the following two projects.

This website

TODO say more about this project:

Stan - a probabilistic programming language

Stan is software targeting Bayesian statistics and, more generally, high-performance statistical computing. The core development team is a friendly and active group of computer programmers turned statisticians -- or is it the other way around? Lately I've been thinking hard about the Hamiltonian Monte Carlo algorithm behind Stan.

Recently, I worked with Bob Carpenter and Brian Ward to develop BridgeStan. Check out the Journal of Open Source Software paper BridgeStan: Efficient in-memory access to the methods of a Stan model.

Sometime ago, I played around with rewriting parts of Stan's internals, and wrote up a technical report despite my failed attempts to make the software faster for all models of interest.