I am a Research Scientist at Google DeepMind working on LLM agents and their learning dynamics, with a particular focus on reinforcement learning (RL), and a broader goal of helping build universal AI assistants—systems that can understand and remember context, use tools, act on users’ behalf across devices, and, in the longer term, help accelerate scientific discovery.

Before joining Google DeepMind full-time, I completed a PhD in machine learning at the University of Cambridge, supervised by Prof. Mihaela van der Schaar FRS in the Machine Learning and Artificial Intelligence group. My research focused on reinforcement learning for and with large language models: memory-augmented agents, tool use, scalable long-context generation, and automating parts of the scientific pipeline. My first-author work has appeared at NeurIPS, ICLR, ICML, AISTATS, and RSS, including multiple spotlights and a long oral. Earlier, I also worked with Google DeepMind on hierarchical RL and the MuJoCo Playground project.

Some recurring themes in my work are:

• LLM agents with memory, tools, and lookahead
• Unbounded code and content generation (e.g., large codebases, long-form text)
• Agents optimising algorithms and simulators for science
• RL at inference time for adapting language model behaviour

Here on GitHub you’ll mostly find research code and prototypes for these directions, along with open-source tools and experiments developed in my PhD focused on scalable, self-improving agentic systems.