I am a researcher at Google DeepMind working on improving the agentic robustness and tool use and function calling capabilities of Gemini. Before joining GDM, I built and led Cohere’s post-training team from the ground up, shipping the Command Nightly models that topped the HELM leaderboard, Command R+ — a best-in-class open-weight model — and the enterprise-ready flagship Command A. My research has received Best Paper awards at NeurIPS and ACL, and Outstanding Paper at EMNLP. Until 2025, I was co-chair of the Dynabench and the Data-Centric Machine Learning Research (DMLR) working groups at MLCommons. I also developed and taught the MSIN0221 Natural Language Processing module at the UCL SoM.
Previously, I interned at DeepMind with Po-Sen Huang and Johannes Welbl, and I’ve collaborated with Facebook AI Research (FAIR) under the guidance of Douwe Kiela and Robin Jia on dynamic adversarial data collection, improving model robustness and introducing generative assistants.
My PhD, under the supervision of Pontus Stenetorp and Sebastian Riedel with the UCL NLP group focused on the adversarial robustness of Language Models with humans and models in the loop. I have a Masters degree from the UCL Department of Computer Science and a Bachelors in Mechanical Engineering from the University of Malta.
