cv

PhD, Computer Science

• Advised by Dr Ali Butt, a professor at Virginia Tech, leading Distributed Systems and Storage Lab at Virginia Tech

Masters, Computer Science

• Advised by Dr Ali Butt, a professor at Virginia Tech, leading Distributed Systems and Storage Lab at Virginia Tech

Bachelor of Computer Software Engineering

Research Assistant at Distributed System and Storage Lab at Virginia Tech

• \(\small \textbf{SODA-LLM:}\) Led the design and implementation of SODA-LLM, a storage-aware tensor placement system for distributed LLM training (PyTorch Tensor Parallelism, Fully Sharded Data Parallelism, DeepSpeed ZeRO), reduced end-to-end latency by 20% and memory usage by 5% on LLaMA-3-8B and Qwen-2-1.5B
• Extended \(\small \textbf{OmniFed:}\), a federated learning system framework, with support for gradient compression during the Federated gradient aggregation stage on top of a Ray-based cluster, improving cross-site data-transfer efficiency and reducing communication overhead by 20% in federated training at HPC and cloud scale.
• \(\small \textbf{SHADE:}\) (A collaboration between Virginia Tech and UVA), an input data caching system for distributed CV training for cloud-scale infrastructure; increased cache hit ratios by up to 4.5×over LRU and optimized data throughput.

Research Intern

• Led the design of IO-SENSE, a multi-agent AI system predicting I/O pattern shifts driven by application semantic changes with 90% accuracy

Student Assistant at NERSC, Lawrence Berkeley National Laboratory (LBNL), internship

• I investigated the causes of I/O hotspots in HPC applications and analyzed common performance issues. Specifically, I examined Drishti, an HPC I/O recommendation tool, and found it generates many false positive warnings, and derived insights to improve the accuracy of the the Drishti I/O recommendation tool.

Student Assistant at Lawrence Berkeley National Laboratory (LBNL), internship

• Analyzed I/O variability for HPC workloads (E3SM, LAMMPS) on Lustre, identifying bottlenecks in shared access, metadata contention, and OST data distribution.

Instructor

• Taught CS3214, Computer Systems, at Virginia Tech in Fall 2022 and Spring 2023

Internship at Oak Ridge National Laboratory, Analytics & AI Methods at Scale Group.

• I studied and characterized the application I/O pattern using clustering techniques, and then extracted features from their submitter like the job runs of the same application over a time-window for that user, the scale of the job submissions the submitter tends to submit. This work resulted in my ICDCN'25 paper, where I was able to predict the I/O pattern of the next job given the features from the past submissions of the same user with an accuracy of nearly 90% for HPC jobs.

• Was awarded travel support by USENIX for USENIX FAST'24 conference in Santa Clara, CA, US
• Had a poster at IPDPS24 PhD forum about our I/O interference characterization project with Berkeley Lab (LBNL). I was also awarded travel support by TCPP for IPDPS24.

• Being appointed as the student volunteer for SC23 hosted in Denver, CO, US

• Being appointed as the student volunteer for SC22 hosted in Dallas, TX, US