About
Whether I shall turn out to be the hero of my own life, or whether that station will be held by anybody else, these pages must show.
Charles Dickens, David Copperfield
Hi! I’m Dhruv Rawat.
I am currently working as a Quantitative Researcher and Developer at Nomura Global Markets. My work centers on the architecture of low-latency pricing and analytics engines for financial securities, where I focus on optimizing C++ to ensure deterministic correctness and responsiveness under high-throughput conditions. Prior to this, I worked on Change Data Capture (CDC) in YugabyteDB at Yugabyte.
I studied Computer Science and Economics at BITS Pilani, where I gravitated toward systems and compilers. Along the way I also picked up a taste for research, spending time in Advanced Data Analytics and Parallel Technologies Lab, mentored by Dr. Jagat Sesh Challa and publishing work on scalable machine learning algorithms for streaming and large-scale data.
My work revolves around the simple philosophy of mechanical sympathy. Great software must respect the hardware it runs on.
These days, my research interests lie at the intersection of distributed systems and machine learning. I am specifically interested in how systems can move beyond static heuristics to employ learning-based control planes that autonomously detect design assumption violations and reconfigure resources in real-time.
I am currently driven by three broad lines of inquiry:
How can systems detect that workload characteristics have drifted from design assumptions early enough?
What mechanisms can enable autonomous reconfiguration, and at which layer should they live?
How can system-level adaptivity be co-designed with application-level execution strategies, particularly for large-scale ML workloads?
Outside of work, I enjoy watching films, and following global politics and world affairs. I am also an avid reader of history and love classical music.
Fun Fact: My Erdös number is 5 via Dr. Poonam Goyal ( → Dr. Bengt Enflo → Dr. Per Enflo → Dr. Andrew Granville → Dr. Paul Erdös).
Visit coursework for more detailed information about the courses I did.
Visit coursework for more detailed information about the courses I did.
-
Jagat Sesh Challa, Dhruv Rawat, Navneet Goyal, & Poonam Goyal AnyStreamKM: Anytime k-medoids Clustering for Streaming Data. 2022 IEEE International Conference on Big Data (Big Data), Osaka, Japan.In this paper, we present AnyStreamKM, a framework for anytime k-medoids clustering of data streams. It employs a hierarchical data indexing structure, AnyKMTree, which organizes incoming stream data into a hierarchy of micro-clusters. It supports anytime processing while filtering noise and outliers. Experimental results show that AnyKMTree produces more compact and purer micro-clusters, and when combined with offline k-medoids clustering such as PAM (Partitioning Around Medoids), yields higher-quality results than state-of-the-art methods.
