CVCV 🎓Google Scholar inLinkedIn

Abhishek Vijaya Kumar

I like to design and build ML systems.

I graduated with a Ph.D. in Computer Science from Cornell University, where I was fortunate to be advised by Rachee Singh. I am currently working on RL infrastructure for post-training LLMs. My recent work in this space looks at reducing overheads of calling external tools during RL post-training and inference. During my PhD, I have also worked on exciting projects related to communication optimization for distributed machine learning. My work covers both electrical interconnects like Nvlinks and emerging silicon photonics interconnects like this.

Before Cornell, I spent two amazing years in the Systems group at Microsoft Research India, advised by Muthian Sivathanu and Saikat Guha. I received my undergraduate degree from PES University, Bengaluru.

News

Jul 2026

PCCL will appear at NSDI 2027.

May 2026

TVCACHE will appear at ICML 2026.

Jan 2026

Reconfigurable Torus Fabrics for Multi-tenant ML will appear at ASPLOS 2026.

Spring 2024

Awarded the Bowers CIS-LinkedIn grant for the 2024-2025 academic year.

Publications

TVCACHE: A Stateful Tool-Value Cache for Post-Training LLM Agents

Abhishek Vijaya Kumar*, Bhaskar Kataria*, Byungsoo Oh, Emaad Manzoor, Rachee Singh (*co-first authors)ICML 2026

Paper

PCCL: Photonic Circuit-Switched Collective Communication for Distributed ML

Abhishek Vijaya Kumar, Arjun Devraj, Rachee SinghNSDI 2027

Paper

Reconfigurable Torus Fabrics for Multi-tenant ML

Abhishek Vijaya Kumar, Eric Ding, Arjun Devraj, Darius Bunandar, Rachee SinghASPLOS 2026

Paper

Aqua: Network-Accelerated Memory Offloading for LLMs in Scale-Up GPU Domains

Abhishek Vijaya Kumar, Gianni Antichi, Rachee SinghASPLOS 2025

Paper BibTeX

Chip-to-chip photonic connectivity in multi-accelerator servers for ML

Abhishek Vijaya Kumar, Arjun Devraj, Darius Bunandar, Rachee SinghOFC 2025

Paper BibTeX

A case for server-scale photonic connectivity

Abhishek Vijaya Kumar, Arjun Devraj, Darius Bunandar, Rachee SinghHotNets 2024

Paper BibTeX

Chisel: An optical slice of the wide-area network

Abhishek Vijaya Kumar, Bill Owens, Nikolaj Bjoner, Binbin Guan, Yawei Yin, Victor Bahl, Rachee SinghNSDI 2024

Paper BibTeX

Quiver: An Informed Storage Cache for Deep Learning

Abhishek Vijaya Kumar, Muthian SivathanuFAST 2020

Paper BibTeX

Efficient AllReduce with StragglersPREPRINT

Arjun Devraj, Eric Ding, Abhishek Vijaya Kumar, Robert Kleinberg, Rachee SingharXiv 2025

Paper

Contact

Please feel free to contact me or drop by my office if you are at Cornell.

Email: abhishek [at] cs [dot] cornell [dot] edu
Office: 456, New Bowers CIS building, Hoy Road, Ithaca, NY, 14853 📍

Miscellaneous: In my free time, I like to run 5Ks on this track, eat chaat with friends and play RPG games on PS5. My personal favorite is Dragon Ball Z sparking zero.