MS Computer Science and Engineering
University of Michigan, Ann Arbor
Hi! I'm Dhruv. I am a Masters student in Computer Science at the University of Michigan, Ann Arbor. My research interests sit at the seam of distributed systems and machine learning. Previously, I have worked as a software engineer at Yugabyte and as a quantitative researcher at Nomura Global Markets.
I studied Computer Science and Economics at BITS Pilani, where I spent two years in the Advanced Data Analytics and Parallel Technologies (ADAPT) Lab, mentored by Dr. Jagat Sesh Challa.
I previously worked on anytime stream clustering algorithms, and my industry work has been on distributed storage internals and low-latency systems. Across both, performance was determined by memory layout, data movement, and arrival patterns rather than by asymptotic cost. I am now interested in machine learning systems, where the same constraints appear at larger scale and interact with model quality directly. My interests span four areas:
Parallelism strategies, collective communication, GPU cluster scheduling, elasticity under preemption, fault tolerance and checkpointing, straggler mitigation, energy efficiency
Continuous batching, KV cache management and compression, chunked prefill, prefill-decode disaggregation, speculative decoding, request scheduling under latency SLOs, GPU multiplexing
Tensor compilers and auto-scheduling, kernel generation and fusion, post-training quantization, one-shot pruning, structured sparsity, mixed-precision kernels, algorithm-hardware co-design
Agentic serving, model routing and cascades, semantic and prefix caching, declarative pipeline abstractions, end-to-end cost-quality optimization, evaluation beyond single-model benchmarks
I am looking for research experience and plan to apply to PhD programs. If you work on any of this, I would be glad to hear from you.
We present AnyKMTree, an R-tree adaptation using 2-PAM for node splitting, which supports anytime k-medoids clustering over stochastic data streams. An adaptive noise-handling mechanism detects concept drift cheaply, and the resulting clusterings improve on purity and silhouette against state-of-the-art baselines such as ClusTree and LiarTree.
Master of Science (M.S.) in Computer Science and Engineering
Relevant Coursework
Master of Science (M.Sc.) in Economics
Bachelor of Engineering (B.E.) in Computer Science
GPA: 9.2/10 · Distinction · Institute Merit Scholar
Relevant Coursework
The best way to reach me is via email. The following address will get to me:
dhruvr [at] umich [dot] edu
Header photograph by Vaibhav Kaul (@Himalayologist).