Abstract

In this talk, Pascal Kesseli explores how formal methods can enhance reasoning capabilities in modern AI systems, drawing on his research at the University of Oxford and his work at Meta GenAI and FAIR. He introduces an agentic inference-time reasoning engine, Logic.py, and shows how it supports structured decision-making in agents. He then turns to follow-on work that applies AlphaGo-style reinforcement learning training pipelines to improve reasoning behavior, and he closes by outlining how inference-time and training-time approaches are beginning to converge into a unified direction for building more capable, dependable agents.

Speaker Bio

Pascal Kesseli is a researcher and engineer working at the intersection of formal methods and AI. He earned a DPhil from the University of Oxford, focusing on static code analysis and program synthesis using formal reasoning engines. At Meta, he worked across GenAI and FAIR on agentic LLM systems and research infrastructure, including inference-time tool use with formal reasoning engines, contributions to Meta’s Code World Model, and reinforcement-learning environments grounded in static analysis. Most recently he started a position at Microsoft AI, working on RL and coding capabilities.

Pascal Kesseli is a researcher and engineer working at the intersection of formal methods and AI. He earned a DPhil from the University of Oxford, focusing on static code analysis and program synthesis using formal reasoning engines. At Meta, he worked across GenAI and FAIR on agentic LLM systems and research infrastructure, including inference-time tool use with formal reasoning engines, contributions to Meta’s Code World Model, and reinforcement-learning environments grounded in static analysis. Most recently he started a position at Microsoft AI, working on RL and coding capabilities.