Computational Physicist • Machine Learning Engineer • Research Software Engineer • Scientific Computing

I develop high-performance Machine Learning and scientific computing software for computational physics research, with expertise in PyTorch, JAX, CUDA, and large-scale GPU-accelerated simulation.

My work spans Machine Learning research, HPC, GPU acceleration, optimization algorithms, and scientific software engineering.

• PyTorch Pace-Setter Award 2023 — awarded by the PyTorch Foundation for excellence in high-level contributions to the PyTorch ecosystem
• 10,000× System Scaling & Optimization — Co-authored a high-performance simulation package to estimate the quantum resources of wavefunctions in the computational basis, leading the native GPU-acceleration architecture; developed custom CUDA.jl kernels that expanded supported system sizes by 10,000× and slashed execution walltimes from months to hours. (Status: Codebase currently proprietary; scheduled for open-source release upon publication of the corresponding research article)
• First application of ML to nuclear structure calculations — published in peer-reviewed journals and presented at international conferences across Europe and the US
• 170+ citations | h-index 6 | i10-index 3

Machine Learning

Languages

Systems & Dev

High-performance optimization framework for Neural Quantum States (NQS).

• Implemented scalable second-order optimization strategies including KFAC and Stochastic Reconfiguration.
• Engineered a novel optimization strategy called Decisional Gradient Descent (motivated by Game Theory), which improves numerical optimization over Stochastic Reconfiguration.

Companion code for a published paper: Drissi, M., et al. "Second-order optimization strategies for neural network quantum states." Philosophical Transactions A 382.2275 (2024): 20240057.

The first open-source application of Machine Learning to nuclear structure calculations.

• Engineered the first application of neural networks in representing nuclear many-body systems - the deuteron.
• Demonstrated that neural networks can represent nuclear wavefunctions to high degrees of accuracy.

Companion code for the published paper: Keeble, J. W. T., and A. Rios. "Machine learning the deuteron." Physics Letters B 809 (2020): 135743.

• Keeble, Lovato, Robin — "Neural Quantum States in Non-Stabilizer Regimes: Benchmarks with Atomic Nuclei" — arXiv:2603.28646 (2026)
• Drissi, Keeble, Rozalén Sarmiento, Rios — "Second-order optimization strategies for neural network quantum states" — Phil. Trans. R. Soc. A 382, 20240057 (2024)

Open to ML Engineering, Research Engineering, and Scientific Computing roles — particularly in ML/AI, HPC, or quantitative research.