I am a PhD student in Computer Science at UT Austin, where I am fortunate to be advised by Prof. Purnamrita Sarkar and Prof. Kevin Tian. Previously, I completed my MS at UT Austin and my BTech with Honours at IIT Bombay, both in Computer Science. At IIT Bombay, I was fortunate to be advised by Prof. Suyash Awate on medical imaging and to work with Prof. Preethi Jyothi on natural language processing for Hindi-English code-switched text.
I work on theoretical machine learning and statistics, particularly high-dimensional estimation, sampling, and generative modeling. My recent work includes streaming and sparse PCA, differential privacy, diffusion models, and planning with language models.
I have also been fortunate to be mentored by Jiang Guo at Amazon and Dheeraj Nagaraj at Google DeepMind. In summer 2026, I worked with Jiang in the AWS Agentic AI team on diffusion language models for planning and plan repair. In summer 2024, I worked with Dheeraj on score matching for diffusion models.
I spent summer 2022 at Microsoft Bing as an applied scientist intern, working on dense retrieval for search. Before graduate school, I was a software engineer at Google Maps, where I worked on detecting abuse in user contributions.
Preprint, 2026
PaperWe use diffusion language models to generate structured plans in parallel and repair selected steps by masked infilling. Experiments study plan generation speed and repair success in long-horizon tasks.
Preprint, 2026
PaperWe learn transport maps between Markov trajectory distributions using conditional flow matching, with finite-sample guarantees under mixing assumptions and applications to EEG domain adaptation.
Preprint, 2026
PaperWe give near-optimal guarantees for Oja's algorithm without an eigengap assumption or rank-one updates, and apply them to differentially private PCA.
Preprint, 2026
PaperWe exploit sparsity to sample from fixed-magnetization Ising models at low temperatures and reduce the measurement complexity of Bayesian sparse linear regression.
Preprint, 2026
PaperWe use graph random walks to study how conditional dependence affects parallel sampling in masked diffusion models, and develop bisection samplers that are exact for random walks with exact learned marginals.
NeurIPS 2026 (Spotlight)
PaperWe adapt leverage score sampling to the geometry of the constraint set, reducing the number of label queries needed for constrained linear regression.
UAI 2026
PaperWe allocate a fixed sample budget across multiple distributions to estimate their means, with nearly optimal regret guarantees and matching lower bounds in several distribution classes.
COLT 2026
PaperWe establish dimension-dependent lower bounds for private sparse covariance estimation and PCA, and show how additional eigenvector sparsity permits dimension-efficient private PCA.
ICML 2026
PaperWe analyze a variant of the truncated power method for sparse PCA with general covariance matrices, obtaining global convergence without a good initialization.
NeurIPS 2025
PaperWe establish nearly dimension-free sample complexity bounds for score learning, apart from hypothesis-class complexity, and introduce bootstrapped score matching to reduce variance across noise levels.
NeurIPS 2025
PaperWe study Oja's algorithm under stochastic quantization and show that batched variants attain quantization lower bounds up to logarithmic factors.
NeurIPS 2025
PaperWe give a nearly-linear-time differentially private algorithm for computing the geometric median.
COLT 2025
PaperWe give provable algorithms for spike-and-slab posterior sampling at any signal-to-noise ratio, using a number of measurements sublinear in the problem dimension in sparse regimes.
UAI 2025
PaperWe construct confidence intervals for individual entries of the eigenvector estimated by Oja's streaming PCA algorithm.
NeurIPS 2024
PaperWe analyze Oja's algorithm with thresholding for streaming sparse PCA, achieving near-optimal rates of convergence.
NeurIPS 2024
PaperWe develop a nonparametric score to select among independent component analysis methods in the presence of arbitrary Gaussian noise.
COLT 2024
PaperWe analyze deflation-based algorithms for k-PCA using 1-PCA oracles.
NeurIPS 2023 (Spotlight)
PaperWe give a nearly optimal streaming algorithm for PCA when samples have Markovian dependence.
ACL-IJCNLP 2021 (Oral presentation)
PaperWe adapt neural machine translation models to generate Hindi-English code-switched sentences from monolingual Hindi sentences.
SPIE Medical Imaging 2019 (Oral presentation)
PaperWe develop an Android application for three-dimensional scanning of surface wounds and compare open-source reconstruction libraries.
* Equal contribution.
For a complete list of projects, please refer to my CV.
I worked at Google, Bangalore with the Maps team on a model for predicting the correctness of an edit made by a user on Maps. This model is used to ensure that data displayed on Google Maps is true to the accuracy it claims.
Implemented an AI for playing Chinese Checkers using the Minimax algorithm with alpha-beta pruning. (Implemented in Racket)
Graduate Teaching Assistant, CS 361S, Network Security and Privacy, UT Austin, Fall 2021.
Graduate Teaching Assistant, EE 461P, Data Science Principles, UT Austin, Spring 2022.
Teaching Assistant, Linear Algebra for Machine Learning, Sandbox@Alphabet.
Undergraduate Teaching Assistant, PH107, Quantum Physics and its Applications, IIT Bombay, Fall 2017.