About me

I am a Postdoctoral Researcher in arithmetic geometry at ETH Zürich. My doctoral advisor was Richard Pink. My research focuses on the stable reduction of hyperelliptic curves, and I am currently thinking about applications to the Parity conjecture. For the SageMath algorithm calculating our classification for genus 2 curves, see here.

AI for Mathematics

I am interested in the interaction between AI and research-level mathematics, especially in evaluation settings that reflect mathematical practice beyond final-answer problems. I am part of the organizing team of IMProofBench, a private proof-based research-level mathematics benchmark.

Furthermore, I am part of the development team of ProofCouncil, an open-source LLM agent for tackling open mathematical problems. With this system, we took part in the FirstProof: Second Batch competition.

I am also involved with MathArena, where I work on ArXivMath, BrokenArXiv, and ArXivLean.

Finally, I have contributed questions to multiple research-level datasets such as Humanity's Last Exam, and I am interested in how AI will affect the mathematical community.

List of publications

Published

• Reduction of Hyperelliptic Curves in Residue Characteristic 2 (with R. Pink), Journal of Number Theory 281 (2026), 429-491. DOI:10.1016/j.jnt.2025.10.004 arXiv:2404.14214
• A benchmark of expert-level academic questions to assess AI capabilities (Humanity's Last Exam, HLE Contributors Consortium), Nature 649 (2026), 1139-1146. DOI:10.1038/s41586-025-09962-4 arXiv:2501.14249

To Appear

• The Classification of the Stable Marked Reduction of Genus 2 Curves in Residue Characteristic 2, to appear in Mathematics of Computation. arXiv:2507.02426

In Review

• Computing the Stable Reduction of Hyperelliptic Curves in Residue Characteristic 2, 2025. arXiv:2506.19663

Preprints

• Benchmarks in Leipzig (with A. Balakin, M. Bóna, M.-C. Brandenburg, et al.), 2026. arXiv:2606.05818
• Beyond Benchmarks: MathArena as an Evaluation Platform for Mathematics with LLMs (with J. Dekoninck, N. Jovanović, K. Rögnvaldsson, I. Petrov, C. Sun, and M. Vechev), 2026. arXiv:2605.00674
• IMProofBench: Benchmarking AI on Research-Level Mathematical Proof Generation (with J. Schmitt, G. Bérczi, J. Dekoninck, J. Feusi, et al.), 2025. arXiv:2509.26076
• Reduction of Hyperelliptic Curves in Characteristic ≠ 2 (with R. Pink), 2021. arXiv:2112.05550

For Humanity's Last Exam, my contribution comprised 12 questions, mostly in mathematics. The published benchmark contains 2,500 expert-level questions across many academic disciplines, developed by nearly 1,000 subject experts.