about.

I'm an incoming PhD student at Stanford University studying AI and formal methods under Clark Barrett. Earlier this spring, I completed my master's in Mathematics and bachelor's degrees in Mathematics and Computer Science at Carnegie Mellon University.

My primary line of research is on machine learning for mathematics, and more generally, AI for science. I also study higher category theory, homotopical algebra, and type theory. I worked under Prof. Jeremy Avigad, Prof. Sean Welleck, and Prof. Prasad Tetali as a researcher in the Hoskinson Center for Formal Mathematics, the Language Technologies Institute, the L3 Lab, and the Institute for Computer-Aided Reasoning in Mathematics. I am the cofounder of Stagira Labs, a research lab focused on using game theory and multi-agent RL to scale AI for scientific discovery. I am also part of the Homotopy Type Theory (HoTT) group and the Higher Category Theory reading group at CMU. Previously, I also worked with Prof. Po-Shen Loh on scaling AI-assisted theorem proving on math competition problems at Expii, and ran The Purple Hand, an anti human trafficking nonprofit.

Recently, I also won the Richard A. Moore Award and was named an ACS Scholar.

My Erdos number is 2.

breaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking news
[Apr 20, 2026]

Defended my Master's!

[Mar 18, 2026]

Accepted an offer to work on AWS Trainium and Neuron compiler

[Mar 15, 2026]

Accepted a PhD position at Stanford

[Mar 1, 2026]

Multi-agent theorem proving paper released

[Feb 14, 2026]

Inducted into Phi Beta Kappa

[Jan 29, 2026]

ImProver2 released

[Nov 30, 2025]

cLean milestone report released

[Sep 23, 2025]

Named as an ACS Scholar

breaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking newsbreaking news

research.

My research focuses on automated theorem proving and machine learning for formal mathematics. I'm particularly interested in developing AI systems that can assist mathematicians in discovering and verifying theorems in interactive proof assistants like Lean 4.

My work spans reinforcement learning for proof search, proof optimization techniques, and building tools that make formal verification more accessible to researchers and practitioners. I believe that AI-assisted theorem proving can accelerate mathematical discovery and improve software verification.

ImProver: Agent-Based Automated Proof Optimization

Riyaz Ahuja, Jeremy Avigad, Prasad Tetali, Sean Welleck

ICLR 2025

Agentic proof optimization framework for Lean for the rewriting of undergraduate and competition-level theorems.

ImProver 2: Iteratively Self-Improving LM's for Neurosymbolic Proof Optimization

Riyaz Ahuja, Tate Rowney, Jeremy Avigad, Sean Welleck

In Submission (ICML 2026)

Generalized RL Pipeline for training proof optimizers for the optimization of research-level Lean theorems at scale, across metrics, and at low cost.

DSLean: A Framework for Type-Correct Interoperability Between Lean 4 and External DSLs

Tate Rowney, Riyaz Ahuja, Jeremy Avigad, Sean Welleck

In Submission (ITP 2026)

A framework for bidirectional translation between the Lean proof assistant and external DSLs for interactive computating in formal theorem proving.

Neural Proof Optimization

Riyaz Ahuja

Master's Thesis on Neural Proof Optimization.

Multi-Agent Theorem Proving

Riyaz Ahuja, Shivansh Gour

Multi-agent reinforcement learning and game theory for decentralized theorem proving.

Unity: Multi-Agent Collaborative Autoformalization

Shivansh Gour, Riyaz Ahuja, Jeremy Avigad, Sean Welleck

In Progress

Multi-agent, low-cost, collaborative autoformalization platform.

cLean: Verifiable GPU Programming in Lean

Riyaz Ahuja

In Progress

Framework for writing, executing, and verifying the safety and correctness of GPU programs in Lean.

Agora: Market-Based Multi-Agent Automated Mathematical Discovery

Riyaz Ahuja, Alexander Heckett, Shivansh Gour, Alexander Willoughby, Tate Rowney, Ishin Shah, Chris Su

In Progress

Market-based multi-agent mathematical discovery platform for collaborative human-AI research in formal mathematics at scale.

projects.

When to trust AlphaFold

Analysis and reproduction of AlphaFold2 predictions

CohesiveGalois

Proof and formalization that the Shape functor in Cohesive ∞-Topoi is not conservative in Lean

LeanM2

Type bridge and tactics to connect Macaulay2 to Lean for computational algebraic geometry

Whitney-Graustein Theorem

Formalization of the Whitney-Graustein theorem in Lean

Baire Category Theorem

Formalization of the Baire Category theorem and applications in Lean

Keraunos

Autonomous drone swarm navigation and targeting system

Orbisol

Orbital debris tracking and collision avoidance software

other.

talks.

Language Technologies Institute

Multi-Agent Theorem Proving

L3

ImProver2

Categorical Logic group

Kripke-Joyal Semantics

Algebraic Geometry group

LeanM2

L3

How to build a (useful) theorem prover

L3

ImProver

Differential topology group

Sard's theorem

past work.

Expii

Edtech platform for gamifying interactive math learning

The Purple Hand

501(c)(3) nonprofit for using data to fight human trafficking