News

Our paper, “Towards Network-Efficient Cross-Regional Inference via Learned Activation Compression,” was accepted into the NAIC 2026 workshop, co-located with ACM SIGCOMM 2026 in Denver, CO.

Our poster, “mrLLM: Fast Multi-Region LLM Inference using Learned Adaptors,” was accepted to the NSDI 2026 poster session in Renton, WA.

I gave an invited talk on OptiNIC at the OCP TAP (Open Compute Project, Time Appliances Project).

I will be serving on the Program Committees for APNet 2026, ICNP 2026, MICRO 2026, and NetFabAI 2026. I also served on an NSF CORE review panel.

I presented OptiNIC at the DARPA/SRC ACE Center for Evolvable Computing Industry Liaison Meeting.

I successfully defended my Ph.D. in Computer Science at Purdue University, concluding my doctoral work on systems, networks, and AI/ML infrastructure.

I received the Broadcom Research Award for my work in systems and networking for large-scale AI.

Our paper, “Reimagining RDMA Through the Lens of ML,” was accepted to IEEE CAL 2025 (IEEE Computer Architecture Letters), presenting a new perspective on RDMA architectures for ML workloads.

I presented OptiReduce / Ultima at the DARPA/SRC CUBIC Center for Ubiquitous Connectivity as an invited research presentation.

Visiting Scholar in the Department of Computer Science and Engineering at the University of Michigan, Ann Arbor.

I presented OptiReduce / Ultima at the DARPA/SRC ACE Center for Evolvable Computing Industry Liaison Meeting.

I presented my work on ultra–low-latency CDN edge applications to Microsoft Research (Intelligent Networked Systems Group) and the Microsoft Azure Front Door team.

I joined Microsoft Research in Summer 2025, working on ultra–low-latency applications for Azure Front Door (CDN edge).

The OptiReduce project website is live at optireduce.github.io.

OptiReduce introduces a resilient AllReduce framework that uses bounded-loss reliability to improve tail performance for distributed deep learning. I presented it at NSDI 2025 in Philadelphia, PA.

I was selected to serve on the NSDI 2025 pre-review task force and contribute to the review process.

I received the Google Research Scholar Award for my work on tail-optimal collective communication for distributed training.

I presented Ultima at the IBM/IEEE AI Compute Symposium (AICS ’23) at the IBM T. J. Watson Research Center in Yorktown Heights, NY.

I presented a live demo of Ultima at the FABRIC KNIT 7 community workshop (Sept 27, 2023) and shared early results on resilient collective communication.

I gave a research talk at HPE’s Networking and Distributed Systems Lab on LLM parallelism strategies, the resulting communication patterns, and deployment-aware design choices.

I presented an analysis of RoCE vs. InfiniBand performance gaps for large-scale ML workloads to NVIDIA’s Mellanox networking group.

Ultima was accepted to the OSDI 2022 poster session, presenting early ideas on resilient and tail-optimal collective communication for distributed deep learning.

FuzzUSB was presented at IEEE S&P 2022, introducing hybrid stateful fuzzing for USB gadget stacks.

I presented my work on “Constructing the Face of Network Data” at the SIGCOMM 2021 poster session, exploring new methods for network traffic analysis and security.

I started my Ph.D. in Computer Science at Purdue University, focusing on systems, networks, and AI/ML infrastructure research.