Research

My research lies broadly in stochastic optimization and control, with an emphasis on stochastic networks and queueing theory. A central theme of my work is understanding the steady-state behavior and control of complex stochastic systems, particularly in heavy-traffic regimes. We pioneered the notion of multi-scale heavy traffic and established the first product-form limits for generalized Jackson networks under this regime. I have also studied resource allocation in parallel-server systems, developing steady-state approximations that capture server heterogeneity and class-dependent service rates.

Another focus of my research is high-dimensional stochastic control, particularly in dynamic matching systems. I develop tractable approximations and control policies that exploit the underlying stochastic system structure to balance matching value and congestion costs. This work is motivated by large-scale service systems such as ride-hailing and labor platforms, where the underlying control problems can be extremely high-dimensional.

More recently, I have been exploring LLM inference through the lens of stochastic networks. Modern LLM serving systems involve complex interactions among batching, routing, resource allocation, and congestion under stringent latency requirements. I am interested in developing stochastic models and control methods to understand these systems and improve inference efficiency at scale.