Projects

Biomechanical Foundation Models

Work done Janelia

We are building a full-stack “virtual animal” simulation of Drosophila melanogaster — connecting a large-scale connectome model (250 muscles, full nervous system) to a biomechanically accurate body simulated in MuJoCo/MJX. The goal is to train foundation models that can infer neural activity across the connectome from observed behavior, using reinforcement learning and imitation learning. This builds on our 2025 Nature paper on whole-body physics simulation of fruit fly locomotion.

Chromatix

Work done Janelia

Chromatix is an open-source, GPU-accelerated, fully differentiable wave-optics simulation library built in JAX. It provides a standardized framework for computational optics, speeds up simulation 2–22×, and is used in labs worldwide. Co-developed with Diptodip Deb at Janelia. Published in Nature Methods 2026. GitHub · Paper

X-ray Nano-Holotomography Reconstruction

Work done Janelia

Physics-based reconstruction pipelines for XNH imaging, combining phase retrieval, Paganin-based regularization, and large-scale optimization in JAX on multi-GPU infrastructure. A collaboration with Jakob Reichmann at the Francis Crick Institute, building on the differentiable simulation framework developed in Chromatix.

Quantum-Enhanced Neural Exchange-Correlation Functionals

Work done at Pasqal

Integrating quantum models into Kohn-Sham Density Functional Theory to learn neural exchange-correlation functionals — enabling more accurate molecular simulations via quantum-enhanced neural networks. This work resulted in both a publication in Physical Review A (2026) and a PCT patent (PCT/EP2025/059376) filed with the European Patent Office. Collaboration with Schrödinger Inc. Paper

Rearrangement of Individual Atoms in a 2000-site Optical-Tweezer Array

Work done at Pasqal

Demonstrated rearrangement of atoms across a 2000-site cryogenic optical-tweezer array — a key milestone toward large-scale neutral-atom quantum computing. Built the control software for the holographic system running on an Nvidia Jetson embedded device. Published in Physical Review Applied (2024). Paper

DeepMoD

Work done during my PhD

DeepMoD automatically discovers the partial differential equations underlying noisy, sparse spatiotemporal datasets, combining a neural network function approximator with sparse regression. One of the first AI4Science tools of its kind — 187 citations. Published in Journal of Computational Physics 2021. GitHub · Paper