Harshvardhan Mestha
I graduated from BITS Goa, with a bachelors in Electronics and Instrumentation. I will be joining NYU Courant as an M.S. student in Computer Science starting Fall 2027.

I have been working with Dr. Anand Subramoney's group since 2024, on the topic of State Space Models for in-context learning. In 2025, I spent an amazing few months at TU Dresden in Germany, as a SSMP scholar, working on efficient State Space Model based LLMs for Neuromorphic hardware, under the guidance of Prof. David Kappel and Dr. Subramoney, and I currently continue this work. Previously, I also worked as a researcher at APPCAIR, under the guidance of Prof. Ashwin Srinivasan and Dr. Tanmay Verlekar, on explainable agentic protocols and AI for Healthcare.

I also love all things related to sci-fi, space, cars, and aviation. I also draw and you can find some of my art here.


Publications


BlockMamba: Efficient Scalable Structured Sparsity for Mamba
Harshvardhan Mestha, Khaleelulla Khan Nazeer, David Kappel, Anand Subramoney
Accepted at the 2nd Workshop on World Models @ ICLR'26
[openreview]
lrnnx: A library for Linear RNNs
Harshvardhan Mestha*, et al.
Accepted at the Student Research Workshop @ EACL'26
[arXiv:2602.08810] [proceedings]
Multi-Turn Human-LLM Interaction Through the Lens of a Two-Way Intelligibility Protocol
Harshvardhan Mestha, Karan Bania, Shreyas V, Sidong Liu, Ashwin Srinivasan.
Accepted at Multi-turn Interactions in LLMs @ NeurIPS'25
[arXiv:2410.20600] [poster]
Learning in the Recurrent State: Gradient Descent with Linear Recurrent Networks
Yudou Tian, Neeraj Mohan Sushma, Harshvardhan Mestha, Nicolo Colombo, David Kappel, Anand Subramoney
Under review at NeurIPS 2026
[arXiv:2410.11687] [poster]
Exploring the Missing Medical Context in Generated Radiology Reports
Karan Bania*, Harshvardhan Mestha*, Tanmay Tulsidas Verlekar
Accepted at SLM4Health @ AIME’25
[proceedings]
CountCLIP - [Re] Teaching CLIP to Count to Ten
Harshvardhan Mestha, Tejas Agrawal, Karan Bania, Yash Bhisikar
[arXiv:2406.03586]
*=equal contribution
[more projects (ml/app/webdev)]