My research has been focused on developing provably optimal control algorithms for systems with uncertainty. In particular, I am actively working on dynamic pricing, matching, and charging for matching platforms like ride-hailing platforms and EV-based transportation systems. More recently, I am interested in leveraging applied probability and continuous optimization tools for designing data-driven algorithms that learns the system parameters and customer preferences and makes optimal pricing and matching decisions for such matching platforms.
My Ph.D. thesis on designing provably optimal algorithms for matching platforms received several awards: ACM SIGMETRICS Doctoral Dissertation Award 2024 (Winner), GT Sigma Xi Best Ph.D. Thesis Award 2024 (Winner), and INFORMS TSL Dissertation Award Competition 2025 (Finalist). Before that, my research was also awarded INFORMS Transportation Science and Logistics (TSL) Best Student Paper Award 2023 (Finalist) and Stephen S. Lavenberg Best Student Paper Award at IFIP Performance 2021.
Near-Optimal Regret-Queue Length Tradeoff in Online Learning for Two-Sided Markets (NeurIPS 2025): We study a two-sided market, wherein, price-sensitive heterogeneous customers and servers arrive and join their respective queues. A compatible customer-server pair can then be matched by the platform, at which point, they leave the system. Our objective is to design pricing and matching algorithms that maximize the platform’s profit, while maintaining reasonable queue lengths. As the demand and supply curves governing the price-dependent arrival rates may not be known in practice, we design a novel online-learning-based pricing policy and establish its near-optimality. In particular, we prove a tradeoff among three performance metrics: O(T^1-γ) regret,O(T^γ/2) average queue length, and O(T^γ) maximum queue length for γ in (0, 1/6], significantly
improving over existing results. Moreover, barring the permissible range of γ, we show that this trade-off between regret and average queue length is optimal up to logarithmic factors under a class of policies, matching the optimal one which assumes the demand and supply curves to be known. Our proposed policy has two noteworthy features: a dynamic component that optimizes the tradeoff between low regret and small queue lengths; and a probabilistic component that resolves the tension between obtaining useful samples for fast learning and maintaining small queue lengths.
Graph Alignment via Birkhoff Relaxation (NeurIPS 2025, https://arxiv.org/abs/2503.05323): We consider the graph alignment problem, wherein the objective is to find a vertex correspondence between two graphs that maximizes the edge overlap. The graph alignment problem is an instance of the quadratic assignment problem (QAP), known to be NP-hard in the worst case even to approximately solve. In this paper, we analyze Birkhoff relaxation, a tight convex relaxation of QAP, and present theoretical guarantees on its performance when the inputs follow the Gaussian Wigner Model. More specifically, the weighted adjacency matrices are correlated Gaussian Orthogonal Ensemble with correlation 1/(1+σ^2)^0.5. Denote the optimal solutions of the QAP and Birkhoff relaxation by Π and X respectively. We show that ‖X−Π‖=o(n) when σ=o(n^−1) and ‖X−Π‖=Ω(n) when σ=Ω(n^−0.5). Thus, the optimal solution X transitions from a small perturbation of Π for small σ to being well separated from Π as σ becomes larger than n^−0.5. This result allows us to guarantee that simple rounding procedures on X align 1−o(1) fraction of vertices correctly whenever σ=o(n^−1). This condition on σ to ensure the success of the Birkhoff relaxation is state-of-the-art.
I spent six years (2018-24) in the Industrial and Systems Engineering Department at Georgia Tech, working on my Ph.D., mentored by Prof. Siva Theja Maguluri. Then, I spent one year as a postdoctoral candidate at INRIA Paris, mentored by Dr. Laurent Massoulie. Since August 2025, I am an assistant professor in the IOE department at the University of Michigan.