Nabeel Seedat

Hi 👋 I am a Lead Research Scientist at Thomson Reuters Foundational Research, where I am the research lead for LLM Evaluations as part of the training of Thomson-1, Thomson Reuters’ own frontier-class foundation model (~400B parameters).

Most of my work now sits on the evaluation and verification of LLMs and agents — how do we actually know whether a model works? In practice that means decomposing what a good answer means in high-stakes domains, grounding model claims in authoritative evidence, and making rigorous evaluation orders of magnitude cheaper to run 🚀. I also work on how LLMs reason and search over hard problems, and on who they end up aligning to when demands compete. Throughout, I come at this from a data-centric angle: what you train and evaluate on determines what you can actually trust 🦾.

I completed my PhD in Machine Learning at the University of Cambridge, supervised by Prof. Mihaela van der Schaar. I hold a Masters degree from Cornell University working on Bayesian Deep Learning, as well as a Masters from the University of the Witwatersrand (South Africa) working on Signal Processing & ML for Parkinson’s Disease. I also hold a dual-bachelors in Information Engineering & Biomedical Engineering from the University of the Witwatersrand (South Africa).

Previously, my industry experience includes time as an ML Researcher at AstraZeneca working on LLM verification, test-time scaling and uncertainty quantification. Before my PhD, I worked on production ML systems as a Data Scientist working on Computer Vision at Shutterstock (USA) and as an ML Engineer working on NLP at Multichoice (Africa’s largest multimedia company).

Outside the research itself, I was named one of the Mail & Guardian’s Top 200 Young South Africans 🇿🇦

Always happy to chat about research — feel free to reach out!