Leo Nagel

About Me

Hi, I'm Leo Nagel. I'm working on tooling for semiconductors at Dash Crystal Corporation.

On the side, I'm working on Hessian-aware quantization at the Berkeley AI Research Labs (BAIR). I also keep up to date on research on random matrices and algorithms, where I contribute when I spot something that I want to write about.

I'm German and I live in San Francisco, California. I've spent about 8 years of my life in Germany, 10 years in China, and coming up on 2 years in the United States. I enjoy playing chess and hiking up the Berkeley Hills (at night, especially).

Education

University of California, Berkeley

B. Sc. in Electrical Engineering & Computer Sciences

Class of 2025

Achievements

Berkeley Trading Invitational 2024: Placed first out of 130+ participants in this quantitative finance competition.

The Residency: Participated in Chapter 3 during the summer of 2024.

Global Society of Young Physicists: Founded a physics nonprofit that organizes a summer program and publishes a journal in 2022.

Golden Ridge Society: Founded a nonprofit computer science organization where I wrote the world's fastest implementation of bogosort for the Nvidia H100 (by permutes per second per device).

Publications

Metrics on Permutation Families Defined by a Restriction Graph

Co-authored with Danylo Tymoshenko

arXiv:2507.10569 [cs.DM], July 2025

Understanding the metric structure of permutation families is fundamental to combinatorics and has applications in social choice theory, bioinformatics, and coding theory. This work studies permutation families defined by restriction graphs and characterizes when the Kendall-Tau metric achieves its combinatorial upper bound, revealing connections between metric geometry and poset dimension theory.

Quantitative Edge Eigenvector Universality for Random Regular Graphs: Berry-Esseen Bounds with Explicit Constants

arXiv:2507.12502 [math.PR], July 2025

This paper establishes the first quantitative Berry-Esseen bounds for edge eigenvector statistics in random regular graphs. For any d-regular graph on N vertices with fixed d ≥ 3, we prove that the normalized overlap √N⟨q, u₂⟩ converges to a Gaussian distribution with an explicit convergence rate of N^(-1/6+ε). The proof introduces a single-scale comparison method using constrained Dyson Brownian motion, providing explicit constants throughout the analysis.

Contact

If you're interested in collaborating or have any questions (or want to challenge me in chess), feel free to reach out! Always happy to bring more people on for my projects!

Email: leo (at) lnagel (dot) net

LinkedIn: linkedin.com/in/leonhardn