Contact ↗I develop compilers and kernels for LLM inference at AWS. My research background spans quantum control, superconducting device simulation, and quantum machine learning.
Compiler and kernel engineering,
quantum control, and scientific computing.
I develop compilers and kernels for LLM inference at AWS Annapurna Labs, with a focus on performance and reliable execution.
My role combines software engineering, performance analysis, and collaboration across engineering teams.
I developed reinforcement-learning training and evaluation software for quantum control-pulse optimization using Python, JAX, and PyTorch.
My work included simulation environments, reward computation, simulator feedback, and experimental device data. I evaluated behavior under noise and frequency drift.
I built a Python workflow connecting QTCAD simulations, circuit-parameter extraction, and Hamiltonian generation for superconducting hardware.
I profiled computational bottlenecks and checked selected outputs against analytical references.
I contributed to a CERN openlab study of GANs for two b-jet distributions in simulated Higgs signal events, using a modified DijetGAN architecture.
Our report describes classical training with PyTorch and TensorFlow, comparisons with Monte Carlo distributions, and exploratory continuous-variable quantum GANs using PennyLane and Strawberry Fields.
Coauthored CERN openlab report · 2021 ↗My public contributions include tutorial-based optimizer-comparison notebooks and project documentation.
Stabilization and logical operation of GKP qubits in a fully 2D architectureCoauthored conference abstract · APS · 2025 ↗ On-chip bosonic quantum error correction with a parametrically coupled heavy fluxonium control qubitCoauthored conference abstract · APS · 2025 ↗ Modified Layerwise Learning for Data Re-uploading Classifier in High-Energy Physics Event ClassificationCoauthored conference paper · IEEE QCE · 2021 ↗My research and engineering work draws on simulation, numerical validation and performance analysis. I’ve built optimization tools, checked selected numerical outputs against analytical references, and profiled computational bottlenecks.
Software Development Engineer, Toronto. Building high-performance machine learning systems with a focus on inference, compilers, and kernels. Since November 2025.
Multiscale simulation of superconducting quantum devices. Montréal, February–June 2024.
Generative adversarial networks for Higgs boson ttH process simulation. Geneva, June–October 2021.
A guided comparison of encodings, error syndromes, and recovery.
Explore eight teaching models: encode, add an error, measure, then inspect recovery.
Explore eight code examples: repetition, Steane, Shor, surface, bivariate bicycle (qLDPC), a small hypergraph product, four-component cat, and square-lattice GKP. Repetition includes bit-flip and phase-flip encodings with 3, 5, 7, or 9 qubits. The interactive models need JavaScript.
For a CSS code, the X-error syndrome is sZ = HZeX mod 2. Ideal square GKP corrects small displacements |Δq|, |Δp| < √π / 2 when [q,p] = i.
Read the model conventions and references ↗Measure the checks to obtain the syndrome.
Three error patterns. The same Z0 correction. Choose an ending to see what remains on the code space.
Run this example beside qec-models.mjs ↓ and its helper qec-hgp.mjs ↓, plus qec-phase-space.mjs ↓ and qec-marginals.mjs ↓. Use Node.js or import the modules in a browser. These examples use the visualization’s model calculations and share its limits. The parameters reflect the current experiment.