dphil · engineering science · university of oxford
> AI for quantum devices▌
I'm a PhD student at Oxford (Wolfson College), supervised by Natalia Ares and funded by the Laboratory for AI Security Research (LASR). My research explores the intersection of AI and quantum technologies , spanning both the development of new machine learning methods and their application to challenging scientific problems. I am particularly interested in continual learning, reinforcement learning, world models, and physics-informed machine learning for complex dynamical systems . Before my PhD, I completed an integrated MEng in Engineering Science at the University of Oxford (St Hugh's College) , where I studied machine learning, control, robotics, computer vision, dynamical systems, and applied mathematics. Outside of research, I serve as Events Officer for the Oxford Engineering Society, previously represented undergraduates on the Women in Engineering Network, and was Vice President of the St Hugh's College Music Society.
Continual adaptation for black-box LLM agents with reduced catastrophic forgetting. RIZZ partitions long-term memory and uses verifier-guided retrieval to reduce interference between new and existing knowledge, achieving state-of-the-art aggregate performance across five continual learning and agent benchmarks at substantially lower computational cost than existing memory systems.
Physics-informed neural networks for forecasting driven, dissipative, and chaotic quantum systems. The project investigates when structured inductive biases improve generalization over purely data-driven models, and when they do not.
Predictive world models that continually adapt to changing environments while retaining previously learned dynamics. The project studies transferable latent representations and how world models can discover rules that generalize across related environments.
RL agents that navigate quantum-dot voltage landscapes autonomously, replacing expert manual calibration so that tuning scales to multiple spin qubits.
Machine learning methods for inferring latent, time-evolving charge disorder in semiconductor quantum devices. The project combines accelerated simulations, distribution transformers, and self-supervised learning to recover hidden disorder landscapes from streaming measurements in real time.
A Python simulator for multi-qubit systems under flux and drive-pulse control, with closed-loop Bayesian optimization of pulses for single- and two-qubit gate fidelity.