I'm a junior at WPI studying physics, working toward a research career in observational cosmology and weak gravitational lensing. My work sits at the intersection of machine learning, numerical methods, and astrophysics to build the computational tools that turn telescope images into maps of dark matter.

A JAX/Flax neural network for galaxy shear estimation, developed with my advisor Sayan Saha at Northeastern. ShearNet achieves similar accuracy with faster inference than classical NGmix metacalibration on the same galaxy samples. The architecture uses a fork-like CNN that processes galaxy and PSF images on separate branches before fusing them for shear prediction. A paper is in preparation.

Finite Element Mass Map Inversion is a novel FEM-BEM coupled solver for reconstructing projected mass density maps from weak lensing shear fields. Unlike classical Kaiser–Squires inversion (which uses periodic boundary conditions and suffers from the mass-sheet degeneracy), FEMMI solves the lensing Poisson equation on all of $\mathbb{R}^2$ by coupling a P3 finite element interior to a boundary element method that encodes the correct decay at infinity. The forward operator is injective, resolving the mass-sheet degeneracy. MAP reconstruction uses Morozov's discrepancy principle for automatic regularisation selection. Built independently as part of an independent study in computational physics with Dr. William Sanguinet.

Working with Professor Jacqueline McCleary's group at Northeastern, I've processed real observational data from the SuperBIT stratospheric balloon telescope. This includes calibration and coaddition, galaxy shape measurement with NGmix and metacalibration, and Kaiser–Squires mass reconstruction with E/B-mode decomposition on the galaxy cluster PLCKG287+32.9. I also built VisFITS to explore my curiousity of overlaying mass maps with coaddition images.

JAX · Flax · GalSim · NGmix · SciPy · NumPy · Python · SLURM