Sagnik Anupam

CIS PhD @ UPenn

I am a PhD student in Computer and Information Science at the University of Pennsylvania, where I am fortunate to be co-advised by Osbert Bastani and Dinesh Jayaraman. My research interests include improving the reasoning abilities of large multimodal models in domains like math, programming, visual reasoning, and robotics.

Previously, I graduated with a bachelor's degree from Massachusetts Institute of Technology, where I double-majored in computer science and mathematics. There, I was extremely fortunate to work with Armando Solar-Lezama and Omar Costilla-Reyes on neurosymbolic program synthesis and with Stefanie Shattuck-Hufnagel and Jeung-Yoon Elizabeth Choi on acoustic cue detection. I also spent a summer at Imperial College, London working with Cristopher Salvi. You can find out more about my past experiences here.

Profiles: Google Scholar • X (Twitter) • Mastodon • Bluesky • LinkedIn • Github • ORCiD

I am a member of trustML@Penn, the ASSET Center, and the Perception, Action, & Learning Group at the GRASP Laboratory. I'd love to hear from you if you wish to work with me or collaborate with any of the organizations that I am currently a part of. Feel free to reach out to me via email at {sagnikanupam} at {alum} dot {mit} dot {edu}.

Updates

Oct 2025

Our work on importance-weighted group advantage estimation is out on arXiV!

Oct 2025

Our paper BrowserArena: Evaluating LLM Agents on Real-World Web Navigation Tasks. has been accepted to the Workshop on Multi-Turn Interactions in Large Language Models at NeurIPS 2025!

June 2025

Our paper Effective Reinforcement Learning for Reasoning in Language Models has been accepted to the First Workshop on Methods and Opportunities at Small Scale at ICML 2025!

Mar 2025

Our work on LLM Program Optimization has been accepted to the Third Deep Learning for Code Workshop at ICLR 2025!

Mar 2025

I will be presenting our work on LLM Program Optimization at the meeting of Neurosymbolic Programming Expeditions in Computing at MIT CSAIL from March 6-7, 2025.

Jan 2025