Junior Research Fellow at Christ Church, University of Oxford.
I work on AI planning — given where we are and where we want to be, what actions should we take? My research focuses on building planning systems that are both effective and formally grounded, so we can actually trust their answers. Lately, I've been especially interested in what happens when you combine classical planning algorithms with large language models.
I did my PhD at the University of Basel with Malte Helmert, winning the ICAPS Outstanding Dissertation Award (2025).
- Powerlifted — A lifted classical planning system that works directly over PDDL without grounding. (Part of the Levitron winning planner of the IPC 2023 Classical Satisficing Track!)
- LLM-Generated Heuristics for Planning — A Python framework to generate heuristics for planning domains. We also have a leaderboard with results for more recent models.
- htg-domains — A curated collection of hard-to-ground planning benchmarks.
- grounder — A C++ implementation of the classical planning grounder from Fast Downward for computing relaxed-reachable actions and atoms.
- 📄 Paper at NeurIPS 2025 on using LLMs to generate heuristic functions for classical planning (code in Zenodo)
- 📄 Paper at ICAPS 2026 connecting potential heuristics with multilinear polynomials and Fourier analysis
- 🏆 ICAPS Outstanding Dissertation Award (2025) for my thesis Planning with Different Representations
- 🎓 Co-organiser of the LM4Plan Workshop (Language Models for Planning) at ICAPS 2025, 2026 & ICML 2026