Shravan Nayak

Hi! I’m Shravan Nayak. I am a PhD student in computer science at Mila and the University of Montreal supervised by Prof. Aishwarya Agrawal. My research focuses on building inclusive and robust multimodal systems. Currently, my work spans three areas:

Culture & Alignment. I study how AI can recognize and adapt to diverse cultural perspectives. This includes benchmarking cultural understanding in VLMs (CulturalVQA) and text-to-image models (CulturalFrames), studying pluralistic values and alignment in T2I models (LIVS), and tracing value alignment shifts during LLM post-training (Value Drifts).

Agents. I created the CUA-Suite for computer-use agents, which includes the UI-Vision benchmark, the GroundCUA dataset (downloaded over 150k times), the GroundNext series of grounding models, and the VideoCUA dataset. I am also interested in deploying agents for the enterprise (EnterpriseOps-Gym).

Understanding & Improving VLMs. I am curious about the inner workings of VLMs and how understanding them can help improve their robustness and reliability. I have worked on adversarial robustness in VLMs, interpretablity of visual representations in LLMs (LatentLens), and discovering failure modes in VLMs using RL.

In my free time, I love to cook and share my experiences on my blog. I also have a deep interest in history, culture, and archaeology. I find it fascinating to explore the diverse ways of life and the events that have shaped our world.