Research Fellow — MATS Research

• Technical AI safety research fellowship working with Cozmin Ududec of UK AISI
• Project working on understanding emergent misalignment in LLMs using a Bayesian mode selection picture

Research Scientist — Independent

• Independent AI safety research in collaboration with Google DeepMind researcher Jonathan Richens
• Interpreting and steering model internals of a goal-conditioned RL agent to study how the agent represents its world model and goals, drawing from causal inference theory

Machine Learning Scientist — Depop

• Developed ML-powered recommender system with PyTorch relying on a large-scale ETL pipeline
• Gained experience with Spark, Databricks, Airflow, and AWS services

Research Fellow — AI Safety Camp

• Research on the interpretability of maze-solving transformers
• Implemented activation steering to control transformer behaviour and understand internal representations

Data Scientist — Faculty AI

• Intensive data science course covering supervised/unsupervised learning, neural networks, reinforcement learning, and AI safety
• Developed a recommender system for sales lead generation, achieving a 2.5x efficiency increase

PhD Theoretical Physics — University of Nottingham

• Designed and wrote large-scale C code to simulate gravitational collapse of quantum matter
• Analysed and visualised simulation data with Python
• Published 3 original research papers in high-impact physics journals

MSc Theoretical & Mathematical Physics — Imperial College London

BSc Physics — University College London