Yaseen M. Osman MINDS CDT, 2022

Doctoral Researcher, University of Southampton

Yaseen M. Osman is a doctoral researcher working on how to make large language models more efficient without sacrificing the performance that makes them useful.

Models like those powering modern AI assistants are growing in size and computational demand. Running them typically requires large data centre infrastructure, which carries a significant energy cost and raises questions about where user data goes and who has access to it. Yaseen's research is focused on changing that picture by finding ways to deploy these models on personal devices rather than remote servers.

His approach centres on activation patterns: the internal signals that fire within a transformer model as it processes information. By studying these patterns, Yaseen is developing techniques to understand how models work more deeply and to use those insights to reduce the computational resources they require. The methods he works with include hardware and kernel optimisation, quantisation, pruning and active learning for data efficiency, all aimed at maintaining strong performance while reducing the cost of running and updating the model.

The implications are practical. A model that runs on a low-power device does not need to transmit data to a server, which means user information stays local. It can operate without a network connection and can be updated to reflect new information without the kind of large-scale retraining that cloud-based models depend on. That combination of sustainability, privacy and democratisation is at the heart of what this research is working toward.

His outputs include a preprint available on arXiv and written evidence submitted to a UK parliamentary inquiry. Alongside his research, Yaseen has taken part in two public outreach projects: the Talking Robot, and the Talking River, through which he taught neurodivergent home-schooled children how AI and large language models work. Both projects have been presented at public events and are due to feature at the British Science Festival 2026.

He is due to graduate in 2027.