About
I'm an applied mathematician and engineer working on data-driven simulation of physical systems: reduced-order models, data assimilation, and machine learning, built on top of physics-based solvers. Most of what I do is aimed at one thing — making those methods fast and trustworthy enough to actually use, not just publish.
What I do
Mostly: turning research methods into things that hold up outside the paper they came from.
Simulation Systems
I build simulation pipelines that go from a full finite element solver down to a low-dimensional approximation, without throwing away the physics, so calibration and design loops that used to take hours run in seconds.
Forecasting Pipelines
Kalman-type filters, variational methods, multi-fidelity algorithms — the common thread is combining sparse, noisy observations with a physical model to get a prediction you can actually put a confidence interval on.
Scientific Software
Testing, profiling, documentation, a sane API — I don't bolt these on at the end. They're the difference between a method that works on my laptop and one someone else can pick up and use.
Compute Strategy
MPI, GPU-aware code, distributed snapshot generation, Slurm deployment — the usual toolkit for making sure a method that works on a toy problem still works at the scale people actually care about.
Current work
Right now that means multiphysics and uncertainty pipelines for offshore floating solar — the kind of thing that feeds directly into environmental risk assessment and design decisions, not just a research demo.
Selected Work
GridapROMs.jl
Handles nonlinear, transient, and multi-field problems, with an eye toward reduced models that hold up in a real workflow rather than a single benchmark.
Opals.jl
Kalman, particle, and variational filters behind one interface, with add-ons for localization, inflation, RC-based bias correction, and ROM calibration bolted on as needed. Replaces MeteoModels.jl, an earlier package of mine that's no longer maintained.
Highlights
Developing digital-twin and large-scale multiphysics frameworks for offshore floating solar systems, with forecasting and UQ workflows for reliable design and environmental risk assessment.
Built reduced-order methods, tensor compression workflows, unfitted finite element techniques, and scientific software for parameterized PDEs and computational fluid dynamics.
Worked on topology optimization for compliant aerospace mechanisms under manufacturing and stress constraints, connecting numerical methods to engineering requirements.
Notes
Project note
Interfaces, tests, and benchmarks aren't polish — skip them and the code dies with the paper.
Modeling note
A model can compress the data beautifully and still be the wrong tool once it hits the real world. Some thoughts on why.
Forecasting note
Once uncertainty has to drive an actual decision, a filtering pipeline stops being a math exercise and starts being infrastructure.
Practice
Short notes on simulation, software, and forecasting — I add one when I have something worth saying, not on a schedule.