Dr. Krishna Somandepalli

I’m a Senior Research Engineer at Google DeepMind, where I develop AI systems that learn from multimodal, real-world data—integrating vision, audio, language, and structured signals to understand, model, and generate rich human-centered content.

My recent work includes key contributions to:

• Gemini 2.5 – advancing multimodal reasoning, long-context modeling, and agentic capabilities (2025).
• VideoPoet – a universal autoregressive model for text-to-video generation, awarded Best Paper at CVPR 2024.
• Versatile Diffusion – a patented, flexible diffusion transformer with mixture-of-noise-levels for audiovisual generation (NeurIPS 2024).

Previously, I was a researcher at Google Research and received my PhD in Electrical and Computer Engineering from the University of Southern California (USC), advised by Prof. Shri Narayanan at the Signal Analysis and Interpretation Laboratory (SAIL). My research focused on multimodal representation learning, affective computing, and computational media analysis. At SAIL, I led the Media Informatics and Content Analysis group, developing computational narrative understanding systems for media content.

Before USC, I earned my Master's degree from UC Santa Barbara and worked as a research scientist at NYU Langone, developing models to study functional brain networks using fMRI.

Across all my work—from foundational models to recent AI agents—I’m driven by the goal of building AI that is technically robust, meaningful, and useful to society.