Building the AI that runs companies.

Built a compiler to optimize and parallelize arbitrary PyTorch and Jax code.

LaSR: Symbolic Regression with a Learned Concept Library (November 2023 - May 2024)◦ Conference Paper at NeurIPS 2024, AI4Science Workshop Paper at ICML 2024, arxiv.org/pdf/2409.09359◦ Proposed a more efficient method of evolutionary search for symbolic regression that uses LLMs to evolve on concepts concurrently with evolving on programs◦ First author, under my research in Trishul Lab and Prof. Miles CranmerCOSMOS: Neurosymbolic Grounding for Compositional World Models (November 2022 - November 2023)◦ Conference Paper at ICLR 2024, SSLTheory Workshop Paper at NeurIPS 2023, trishullab.github.io/cosmos-web◦ Introduced a state-of-the-art world modeling framework for generalizing to compositions of objects◦ Second author, under my research in Trishul Lab: Prof. Swarat Chaudhuri’s graduate lab at UT Austin

◦ Developed generative graph models to synthetically mimic the transport layer activity of cyber networks◦ Created a novel algorithm to detect out-of-distribution inputs in object detection models using hidden layer activity◦ Constructed a ML model to determine people’s biometric features (age, gender, race) without using facial data