Anil Kag

I am a Senior Research Scientist and the Lead of the Efficient Generative AI team within the Creative Vision group at Snap Research. My research focuses on the intersection of high-fidelity generative modeling and large-scale efficiency—specifically architecting text, image, video, and audio generation models that are both state-of-the-art and deployable at a global scale. Since joining Snap in 2023, my work has transitioned from academic breakthroughs at top-tier venues like CVPR, NeurIPS, ICML, and ICLR to core product features. This includes Snap Video and SnapGen, which power AI Lenses used by millions of Snapchatters daily. My research also powers personalized image generation and lens creation in Easy Lens. My contributions are currently supported by 9 United States patent applications. You can find a thematic breakdown of my research, patents, and publications in this consolidated research overview.

I earned my Ph.D. in Electrical and Computer Engineering from Boston University, where I was a Hariri Graduate Student Fellow in Prof. Venkatesh Saligrama’s lab specializing in constrained learning and efficient neural architectures. Prior to my doctoral studies, I served as a Research Fellow at Microsoft Research India within the Machine Learning & Optimization Group, collaborating with Dr. Manik Varma on Extreme Multi-label Classification (XMC). I hold a B.Tech. in Computer Science from IIT Guwahati. My early professional experience also includes two years as a Software Engineer at Microsoft India, where I optimized core performance for Dynamics CRM.

Opportunities

• Internships: I am always looking for exceptional Ph.D. students for full-time research internships on my team at Snap Research. If you are passionate about pushing the boundaries of efficient GenAI (on-device or server), please reach out.
• Collaborations: I am always open to discussing new frontiers in efficient generative modeling—from elastic architecture design to novel training algorithms. I have listed some of my current perspectives in my research statement.