Rohit Agarwal

PhD Student in Computer Science

Princeton University

I work on mathematics for AI and the alignment of AI agents. I am advised by Elad Hazan and also collaborate with Sanjeev Arora.

Previously, I graduated from UC Berkeley with a B.S. in EECS (Honors) and a minor in Mathematics, with a 4.0 GPA and Highest Honors.

Rohit Agarwal

About

I am a PhD student in the Department of Computer Science at Princeton University. My research lies at the intersection of mathematics and machine learning. Specifically, I am interested in:

  • Building new optimizers grounded in control theory that are state-of-the-art for diverse workloads, including the pretraining of large language models.
  • Using LLMs for assisting in proof-based mathematics and olympiad problems, including efficient inference pipelines for competition math.
  • The alignment and safety of AI agents, particularly understanding misalignment through theoretical frameworks.

Before Princeton, I was at UC Berkeley, where I worked with Venkatesan Guruswami on approximation algorithms and promise CSPs, in the SLICE Lab with Vighnesh Iyer and Bora Nikolic on hardware verification, and at Lawrence Berkeley National Laboratory on muon collider physics.

Awards

  • Princeton First-Year Fellowship
  • EECS Citation Award (#1 out of 720 graduates)
  • Regents' and Chancellor's Scholar (UC Berkeley)
  • Engineering Dean's List (8 semesters)
  • Outstanding GSI & EECS Outstanding TA
  • Berkeley Integration Bee Medalist

Papers

Selected Papers in AI

Escaping the Cognitive Well: Efficient Competition Math with Off-the-Shelf Models
Xingyu Dang*, Rohit Agarwal*, Rodrigo Porto, Anirudh Goyal, Liam H. Fowl*, Sanjeev Arora
* denotes equal contribution
Submitted to NeurIPS 2026
An inference pipeline achieving state-of-the-art on IMO-ProofBench Advanced (67.1%) using off-the-shelf models at ~$31/problem. Introduces the concept of the "Cognitive Well" and addresses it through conjecture extraction and context detachment.
AI Alignment via Incentives and Correction
Rohit Agarwal, Joshua Lin, Mark Braverman, Elad Hazan
Submitted to NeurIPS 2026
Studies AI alignment through law-and-economics models of deterrence. Formalizes solver–auditor interactions as a bilevel optimization problem and proposes a bandit-based procedure for adaptive reward design that maintains useful oversight and reduces hallucinated incorrect attempts.

Other Computer Science

Hardness of Approximation for 2 → 4 Norms through Quantum Reductions
Rohit Agarwal, Axel Li, et al.
Berkeley Technical Report EECS-2023-236 (Class project, CS270/CS191)
Packed Memory Arrays for Dynamic Graphs in the Distributed Memory Setting
Rohit Agarwal, Alec James, Joshua You
Berkeley Technical Report EECS-2023-235 (Class project, CS267)
New Embedded DSLs for Hardware Design and Verification
Rohit Agarwal, Vighnesh Iyer, et al.
Presented at PLARCH 2023 (co-located with PLDI 2023)

Physics

Towards a Muon Collider
Carlotta Accettura, Rohit Agarwal, et al.
European Physical Journal C (2023). Part of the URAP program at Lawrence Berkeley National Lab.
Muon Collider Forum Report
K.M. Black, Rohit Agarwal, et al.
Journal of Instrumentation 19, T02015 (2024). arXiv:2209.01318.
The Muon Collider
Carlotta Accettura, Simon Adrian, Rohit Agarwal, et al.
FERMILAB-PUB-25-0309-AD-PPD-T (2025).
Interim Report for the International Muon Collider Collaboration (IMCC)
Carlotta Accettura, Simon Adrian, Rohit Agarwal, et al.
CERN Yellow Reports: Monographs, Vol. 2 (CERN-2024-002). arXiv:2407.12450 (2024).
MuCol Milestone Report No. 7: Consolidated Parameters
Carlotta Accettura, Rohit Agarwal, et al.
arXiv:2510.27437 (2025).
MuCol Milestone Report No. 5: Preliminary Parameters
Carlotta Accettura, Rohit Agarwal, et al.
arXiv:2411.02966 (2024).
The Muon Collider — Supplementary Report to the European Strategy for Particle Physics 2026 Update
Carlotta Accettura, Simon Adrian, Rohit Agarwal, et al.
Submitted to ESPPU (2025).

Talks & Posters

Parametric Stimulus Generators for Controllable RTL Fuzzing
Rohit Agarwal, Vighnesh Iyer
UC Berkeley SLICE Lab Summer Retreat, 2023
A Packed-Memory Data Structure for Distributed Graph Streaming
Rohit Agarwal, Joshua You
UC Berkeley SLICE Lab Summer Retreat, 2023
Improving Verification Infrastructure for Chisel RTL
Rohit Agarwal, Vighnesh Iyer
UC Berkeley SLICE Lab Winter Retreat, 2023
Using Cluster Shape for Beam-Background Suppression in a Future Muon Collider Experiment
Rohit Agarwal, Elodie Resseguie, Simone Pagan Griso, Paolo Andreetto, Alessio Gianelle
European Physical Society Conference on High-Energy Physics (EPS-HEP), 2021

Teaching

I love teaching and have been fortunate to be involved in it extensively. At Berkeley, I received an average student rating of 4.95/5 (0.64 above the departmental average) and was recognized with the Outstanding GSI and EECS Outstanding TA awards.

AddisCoder

Teaching Assistant
Addis Ababa, Ethiopia

Helped Ethiopian high school students learn to code as part of the AddisCoder program. Designed labs, exams, and helped over 50 students complete programming assignments in algorithms.

Princeton University

Semester Role Course Title
Spring 2026 TA COS 511 Theoretical Machine Learning (Elad Hazan)
Fall 2025 TA COS 521 Advanced Algorithm Design (Huacheng Yu)

UC Berkeley EECS Department

Undergraduate Student Instructor (uGSI), 2021–2024

Over 3 years, I served on course staff for 10 appointments across 7 unique courses, hosting discussions for a cumulative 300+ students, writing exam questions and homework assignments, and holding office hours and review sessions.

Semester Role Course Title
Summer 2024 TA CS 70 Discrete Mathematics & Probability Theory
Fall 2023 TA EE 120 Signals and Systems
Summer 2023 TA CS 70 Discrete Mathematics & Probability Theory
Spring 2023 TA CS 189 Introduction to Statistical Machine Learning
Fall 2022 TA EECS 126 Probability & Random Processes
Summer 2022 TA CS 61C Great Ideas in Computer Architecture
Spring 2022 TA EECS 16B Designing Information Systems & Devices II
Spring 2022 Reader EECS 126 Probability & Random Processes
Fall 2021 Tutor EECS 16B Designing Information Systems & Devices II
Summer 2021 Tutor CS 61A Structure and Interpretation of Computer Programs

CS Mentors

Junior & Content Mentor
Aug 2020 – May 2021

Ran weekly volunteer sections of 5–6 college students, teaching data structures and discrete math.

Math Outreach & Competitions

I have been deeply involved in mathematical outreach, writing problems and organizing competitions for high school and college students across the United States and internationally, including multiple math competitions in China.

Princeton University Mathematics Competition (PUMaC)

Co-Director & Co-Problem Czar

Co-directing the competition and leading the problem-writing team for 450+ high school students.

Berkeley Math Tournament (BMT)

Problem Writer

Co-wrote the power rounds on Error-Correcting Codes (BMT 2022) and Zero-Knowledge Proofs (BMT 2023) for Berkeley's premier high school math tournament.

ASDAN Math Tournament (China)

Problem Writer

Co-wrote the power round on Computability (AMT 2023) for ASDAN China's high school math tournament, reaching students across China.

Intercollegiate Mathematics Tournament (ICMT)

Problem Writer

Writing and reviewing problems for the collegiate-level competition.

Johns Hopkins Mathematics Tournament (JHMT)

Problem Writer

Contributing problems for high school contestants.

Princeton Puzzle Hunt

Puzzle Writer

Writing puzzles for Princeton's puzzle hunt events.

California Informatics Competition (CALICO)

Problem Tester

Writing and test-solving programming problems for Berkeley's high school competitive programming contest.

Eta Kappa Nu (HKN)

Officer

Developed the website for Berkeley's EECS honor society and volunteered as a tutor for college students.

Course Notes

Below are notes from courses I took at Princeton and UC Berkeley. These are my personal notes and may contain errors — use at your own risk!

Princeton

COS 511
Theoretical Machine Learning
View Notes
COS 521
Advanced Algorithm Design
View Notes
COS 598
Matrix Concentration and Applications
View Notes

UC Berkeley

CS 170
Efficient Algorithms and Intractable Problems
View Notes
CS 172
Computability and Complexity
View Notes
CS 189
Introduction to Machine Learning
View Notes
CS 191
Quantum Information Science and Technology
View Notes
CS 270
Combinatorial Algorithms and Data Structures
View Notes
CS 294-226
Advanced Topics in Error-Correcting Codes
View Notes
EE 120
Signals and Systems
View Notes
EECS 106A
Introduction to Robotics
View Notes
EECS 126
Probability and Random Processes
View Notes
Math 128A
Numerical Analysis
View Notes
Math 202A
Introduction to Topology and Analysis
View Notes
Math 250A
Groups, Rings, and Fields
View Notes
PHYS 137A
Quantum Mechanics
View Notes

Technical Projects

Score AI

Jun 2023

Harnessed Meta's MusicGen model and OpenAI's GPT-4 to create novel soundtracks for writing and movie screenplay. Built with Django + React in 36 hours at the Berkeley AI Hackathon 2023.

Distributed Packed Memory Array

May 2023

Developed a novel graph streaming data structure for graphs with 10 billion+ nodes distributed across multiple servers. Written in C++ and benchmarked against existing solutions.

Matrix Multiplication Kernel

Mar 2023

Optimized the dgemm routine in C++ using OpenMP, cache blocking, microkernels, and communication-minimizing algorithms.

Functional Fuzzing Generative DSL

Nov 2022

A functional, random generative API for QuickCheck-like hardware fuzzing. Supports monadic composition, recursion, and introspection/correlation of random decisions.

Gitlet

Mar 2021

Implemented a Git-like version control system in Java with file persistence, including support for merging and splitting multiple branches and remotes.

Blog

Occasional thoughts on math, machine learning, and whatever else is on my mind. Read on Substack.

Loading recent posts…