Enabling Hardware-Software Co-Design for Large Scale AI Systems via Fine-Grained GPU Memory Tracing
MemNetAI Workshop @ ACM SIGCOMM, Denver
Advised by Dr. Rinku Shah, in collaboration with Marvell Technology.
New Delhi, India
Systems and networks for distributed AI/ML workloads, collective communication performance optimization, edge–cloud architectures, and transport protocols for HPC/AI applications. I'm also interested in Mixture-of-Experts inference and 6G edge/RAN inferencing workloads. Previously, I worked in full-stack development at Birlasoft and Cognizant, and contributed to the Mifos Initiative through Code for GovTech.
MemNetAI Workshop @ ACM SIGCOMM, Denver
Advised by Dr. Rinku Shah, in collaboration with Marvell Technology.
IEEE Conference
Co-authored with and under the guidance of Dr. J.N. Singh, Galgotias University.
IIIT Delhi · 9.4 SGPA
NESt (Network & Edge Systems) Lab, advised by Dr. Rinku Shah.
Galgotias University · 8.2 CGPA
Research project on Object Detection using CNNs, presented at IEEE Conference (2023).
Ninth Annual Conference on Machine Learning and Systems — Artifact Evaluation
Reviewed 2 papers and evaluated the attached artifacts on Chameleon testbeds.
IIIT Delhi
Graded answer sheets, ran tutorials, and gave remedial lessons.
Birlasoft, part of the Hewlett Packard team
Scalable backend and data architecture projects; designed and optimized RESTful APIs.
The Mifos Initiative (Code for GovTech)
Built an Open Banking fintech app with Java Spring Boot and PostgreSQL, collaborating with teams across India, the UK, and the US.
Cognizant
Built a multithreaded ASP.NET project; served as Cohort Lead during training.
IIIT Allahabad
ML Decision Tree algorithm design and implementation.
Profiled network stack, CPU, and GPU on a 2×RTX / 2×H100 testbed running DDP on ResNet, with and without containerisation, to measure overhead cost.
Comparative study of NCCL, MPI collectives, and UCCL; deep dive into NCCL internals using Nsight Compute, Nsight Systems, and CUPTI. Aug 2025–Present.
Optimization techniques for Mixture-of-Experts inference across DeepSeek, Kimi K2, Llama, and Qwen.
A network proxy that reduces latency in a dynamic network between server and client.
Conduct polls, analyse voting patterns, and showcase results, with authentication and security features.
TypeScript full-stack app for creating and managing customised notes.
Java Spring Boot app built to work with Mojaloop, part of the Mifos Initiative open-source project.
Mobile app for identifying currency and objects, with voice-to-text for emailing.
Always happy to talk systems, networking, or GPU internals — reach out by email or connect on LinkedIn.