Ilias Leontiadis

I am a Staff Research Scientist and Tech Lead at Google (London), where I primarily work on the Agentic capabilities of Gemini, focusing on Agentic memory, harness, and orchestration in collaboration with Google DeepMind (GDM) and CoreML teams.

Prior to joining Google, I was at Meta (London), contributing to the Meta Superintelligence Lab (MSL) within the GenAI organization. At Meta, my focus was on the development and rigorous evaluation of Large Language Models (LLMs), shaping post-training and evaluation strategies for foundational models including Llama 3, Llama 4, and Muse. I specialize in architecting robust AI Judges, Reward Models, and Agentic Evaluation Systems that provide critical signals for training state-of-the-art generative models, ensuring their alignment with human preferences across safety, reasoning, and multimodal capabilities.

Professional Journey & Education

• 2005 – 2009: University College London (UCL), PhD in Computer Science.
  Research focus: Mobile and embedded systems.
• 2009 – 2012: University of Cambridge, Research Fellow.
  Research focus: Embedded AI (mobile and microcontrollers).
• 2013 – 2018: Telefonica Research, Senior Research Scientist and Tech Lead.
  Research focus: Deep Learning, Machine Learning for Networking, Mobile Computing, exploring efficient systems, computation offloading techniques, and network automation.
• 2019 – 2021: Samsung AI Center, Principal Research Scientist and Group Leader.
  Research focus: Edge AI, encompassing Distributed Machine Learning, Federated Learning, and On-device Personalization within resource-constrained environments.
• 2021 – 2026: Meta (Facebook) Superintelligence Lab (MSL), Staff Research Scientist and Tech Lead.
  Research focus: Development of advanced AI Judges, post-training methodologies (reward models), and evaluation strategies for foundational models.
• 2026 – Present: Google, Staff Research Scientist and Tech Lead.
  Research focus: Agentic LLMs and Model Efficiency, focusing on agentic systems (memory, self-improvement, harnesses, orchestration, and evaluation).

Teaching & Service

I am an Affiliated Professor at the Barcelona School of Economics (BSE), where I teach graduate-level courses in Deep Learning and Data Science. I am actively involved in the research community, regularly serving on program committees for top-tier conferences such as NeurIPS, ICLR, ICML, MobiSys, and WWW.