Peyman Hosseini

Queen Mary University of London

Peyman Hosseini is a PhD candidate working on Agentic AI and building on efficient solutions with small language models and post-training of large and small language models to enable these LLMs to be powerful on-device assistants. He has interned for the last 12 months at Samsung Research in the UK where he has led 3 paper sumbissions and 2 patent submissions on post training foundation models with reinforcement learning algorithms as well as building efficient on-device memory agents.

A description of Peyman's research:

Peyman's Rsearch targets post-training of foundation models, specifically large language models, to deliver personalized and powerful AI-powered solutions that are deployable on edge devices, such as mobile phones and personal computers. This is specifically important as Large Language Models (LLMs) are powerful yet impossible to deploy on edge-devices to their computational requirements. On the other hand, Small Language Models (SLMs), i.e., language models between 2-32B params, are more efficient but yet unable to handle complex tasks. Fine-tuning these models to work well in complicated setting enables a lot powerful, privacy-preserving AI-powered applicatios, such as personalized on-device recommendation systems and agents capable of memorizing users' habits and interests.