About me

I am currently Assistant Professor at CSE, IIT Bombay. Until recently I was a post doctoral researcher at sky-lab (previously called RISE), UC Berkeley with Ion Stoica, Joey Gonzalez and Matei Zaharia working on efficient AI systems. Previously, I did my Ph.D. at Rice University advised by Dr. Anshumali Shrivastava in 2024. Here is my thesis.

In a previous life, I used to be strategist at Tower Research Capital where I spent quite some time after graduating from IIT Kanpur in 2013. At IIT Kanpur, I was fortunate to work with Dr. Subhajit Roy, Dr. Mark Marron and Dr. Sumit Gulwani from Microsoft Research on Progamming Language research.

Note for Students

I am looking for students to work on efficient AI research.

For CSE students from IIT Bombay. Drop by my office KR 208. You can mail in advance to ensure I am available. I have floated projects for SURP. So you may want to look at those if you want to work over the summer.

For prospective PG students, once you land at IIT Bombay, we can chat about projects. If you want to get involved before hand look at the note for non-IITB students.

For non IIT Bombay students, look at sky-light which is were I host my research repositories. You can start with implementing some new model families inside sparse attention hub. See for example Gemma Issue, Gemma PR. Once you land a PR, get in touch with me. I would not be able to reply individually to all the student emails. Apologies in advance.

News

🤜 Paper accepted at ICLR’26 : “vAttention: Verified Sparse Attention via Sampling”. paper link

🤜 Paper accepted at ICLR’26 : “vCache: Verified Semantic Prompt Caching”. paper link

⭐ Joined IIT Bombay as Assistant Professor

❣️ Our public page on research is live : https://sky-light.eecs.berkeley.edu/

🤝Serving on Program committee for ICML 2026, ICLR 2026

🤜 Paper accepted at ICML’25 : “HashAttention: Semantic Sparsity for Faster Inference”. Congratulations to the team Shuo Yang, Alejandro Cuadron, Matie Zaharia, Joeseph Gonzalez and Ion Stoica Details coming soon! Heres the arxiv version.

🤜 Paper accepted at ICML’25 : “Sketch to Adapt: Fine-Tunable Sketches for Efficient LLM Adaptation”. Congratulations to the team Tianyi Zhang (lead), Junda Su (co-lead), Oscar Wu, Zhaohuo Xu and Anshumali Shrivastava Details coming soon! Heres the arxiv version.

🤝Serving on Program committee for ICML 2025, Neurips 2025

🤜 Paper accepted at KDD’25 : “IDentity with Locality: An Ideal Hash for Gene Sequence Search”. Congratulations to the team Gaurav Gupta, Tianyi Zhang and Anshumali Shrivastava Details coming soon! Heres the arxiv version.

🤜 Paper accepted at Neurips’24 : “Accelerating Inference with Fast and Expressive Sketch Structured Transform”. Congratulations to the team Kimia Saedi, Apoorv Walia, Keren Zhou, Jihyeong Li and Anshumali Shrivastava Github Page

🤝Serving on Program committee for KDD August 2024, ICLR 2024, AISTATS 2025

❕For highlights of previous news .. refer to the CV