I am a Postdoctoral Researcher at the Waterloo Centre for Astrophysics, University of Waterloo. I am a cosmologist working at the intersection of theory, computation, and statistics to understand the large-scale structure of the Universe and to extract maximal information from current and upcoming cosmological surveys.
My research lines are tightly interconnected: in order to perform extensive cross-validation and field-level analyses of cosmological datasets, fast, differentiable codes are an absolute necessity. This drives the development of the methods I work on:
Accelerating theoretical predictions. I develop and apply numerical techniques to speed up calculations of cosmological observables — including surrogate models and novel algorithms — to reduce computation time without sacrificing scientific accuracy.
Enabling gradient-based inference. I specialize in automatic differentiation methods that power state-of-the-art gradient-based algorithms, both for minimization (L-BFGS) and sampling (Hamiltonian Monte Carlo, Microcanonical Langevin Monte Carlo), enabling scalable and efficient analysis.
Maximizing precision and accuracy in inference. I work on extracting the most possible information from our data through field-level inference techniques, and I rigorously validate results using approaches such as Leave-One-Out and cross-validation to avoid bias and overfitting.
Much of this work is implemented in the Julia programming language, leveraging its performance, composability, and differentiability to unify modeling, optimization, and validation in a single, coherent framework.