Kumar Krishna Agrawal

I am a PhD student at UC Berkeley, affiliated with Berkeley AI Research and UCSF. I am incredibly fortunate to be co-advised by Adam Yala and Trevor Darrell. I also work closely with Sky Computing Lab.

My research is dedicated to building algorithms and systems for human-centric AI.

Advancing Capabilities: How do we build intelligence that assist humans to better perceive, and reason in the real world?
pillar0: new frontier for radiology foundation models
Atlas: multi-scale architecture for long context image-modelling
GanSynth: towards real time music generation (google blog)
Discriminator Actor-Critic: sample-efficient robot learning
AI Systems: How do we design scalable systems for serving long‑context, multimodal models?
vAttention: sparse algorithms for LLM inference
Octopus: towards streaming perception for robots
RedunCut: low-cost live video understanding
Previously