Mathematician · Research Fellow
At King’s College London, I study algebraic groups through the way they act. I use computation and AI to push further into questions that begin in my mathematical research.
01 / about
I am a Postdoctoral Fellow in the geometry group at King’s College London, working on algebraic groups, rational actions, and representation theory.
I completed my PhD at Imperial College London under Martin Liebeck, followed by a year as an INI–Simons Postdoctoral Fellow at the University of Cambridge and two years as an SNSF-funded Research Scientist at EPFL.
Selected talks at Imperial, Manchester, Warwick, Cambridge, EPFL, Banff, and others. Awards include the EPSRC Studentship, Miller Prize, Sanderson Prize, and the Arthur Hinton Read Memorial Prize.
02 / research
I study algebraic groups indirectly, through the way they act on other objects and what their orbit structures reveal. A lot of my work is computational: I write Magma code to understand the actions of matrix groups over finite fields, then connect the results back to the infinite case.
My recent work classifies the generic stabilizers for simple algebraic groups acting on orthogonal and symplectic Grassmannians, and explores multiplicity-free representations and modules with finitely many orbits.
That same computational instinct now drives my work with AI.
03 / mathematics × ai
I use AI to explore mathematical questions I am already pursuing, accelerate computation, and test ideas at scale. I also build custom autonomous AI workflows for search, code, and formalisation. The mathematics sets the direction; AI expands the reach.
// flagship · 2026 · custom Codex workflow
The Monster, the largest sporadic finite simple group, had explicit generators available for every class of maximal subgroup except one: 59:29. We constructed its generators through a fingerprint search that examined 286 million candidates, using a custom Codex workflow to build and run the computation. The construction also yields a short new proof that the Monster has no subgroup PSL₂(59), correcting a 2004 result.
// new paper · 2026 · AI-assisted
We used an AI-assisted falsification search to overturn two conjectures about Saxl graphs. For every base size B ≥ 2, we construct infinitely many primitive groups with a pair of vertices having no common neighbour; at base size two, the failure spans three of the five O’Nan–Scott types. The project combines reproducible GAP, Magma, Python and C++ computations with a Lean 4 formalisation of the all-base-size construction.
// benchmark · 2025
A peer-reviewed benchmark of 77 research-level proof problems written by expert mathematicians, used to evaluate frontier models in an agentic setting, with tools like SageMath and web search in hand.
// next
More mathematical research and custom AI workflows are in progress.
04 / publications
10 papers · 2017–2026
arXiv preprint · math.GR · math.CO
In preparation: a proof of a 40-year-old open problem.
arXiv preprint · cs.CL
Pacific Journal of Mathematics · vol. 336
Forum of Mathematics, Sigma · vol. 13
Journal of Algebra · vol. 606
Journal of Algebra · vol. 570
Acta Mathematica Hungarica · vol. 156
European Journal of Physics · vol. 38, no. 3
05 / side projects
Written up on Medium
//01
Probing the limits of OpenAI’s safety layer for DALL·E 3 using recursive complexity and evasive language.
Read on Medium ↗//02
Gradient-boosted models trained on Lichess data that out-predict the Glicko-2 rating system.
Read on Medium ↗//03
Using the NBA API to find which games across 20 seasons would have flipped with a 30-foot 4-point line.
Read on Medium ↗06 / beyond the math
When I’m not chasing orbits, I play chess, explore London’s theatre scene, and keep up an ongoing (and stubbornly difficult) sourdough journey.
07 / contact