• •Developed networking optimizations to reduce latency in Unity Catalog metadata operations across distributed cloud environments
• •Implemented advanced caching strategies for frequently accessed catalog metadata, improving query performance

• •Worked on low latency trading system for high frequency trading strategies in C++17
• •Implemented order entry and market data feed systems using state-of-the-art technologies like io_uring, eBPF and kernel bypass using Exablaze NIC for exchanges like National Stock Exchange and Chicago Mercantile Exchange
• •Designed and implemented a geographically distributed trading system that executes a unified portfolio strategy across continents and exchanges, utilizing communication channels such as 10Gb Ethernet, microwave, and shared memory

• •Enhanced OLAP cloud warehouse performance by implementing asynchronous SSD scanning with io_uring, significantly increasing concurrent query capacity
• •Optimized C++ system efficiency through rigorous performance benchmarking using Google microbenchmarks, with in-depth analysis via flamegraphs for targeted improvements

• •Developed ONNX to HVX model conversion package for Samsung Neural SDK, optimizing neural networks for Qualcomm AI processors
• •Optimized model loading in Qualcomm Hexagon SDK, reducing on-device AI inference time by 20%

• •Publications in conferences with Prof. Justine Sherry
• •TA for 15-445 Database Systems with Prof. Andy Pavlo

• •Secured rank 982 out of 1 million in IIT JEE Advanced
• •Awarded full tuition waiver for academic excellence

• •Successfully learned the alphabet without eating the crayons
• •Graduated from counting on fingers to actual mathematics

The paper argues that CCA contention between flows no longer dominates bandwidth allocation in today's Internet due to operator policies and application limitations.

The paper presents a novel CCA classifier that uses bottleneck queue occupancy traces and dynamic time warping to accurately identify CCAs with better efficiency and interpretability.

The paper presents a program synthesis pipeline that reverse-engineers CCA from packet traces by formulating synthesis as an optimization problem.