Department of Electrical Engineering
I feel truly fortunate to have had the opportunity to spend this summer as an intern at Ambarella. Over the course of the internship, I worked on improving token efficiency for Video Language Models (Video LMs) by leveraging codec primitives, a project that sits at the intersection of two fields I had not previously combined in my own research: large multimodal models and video compression. Beyond the technical work itself, this internship gave me a first real look into how research is conducted in an industry setting, how a company like Ambarella thinks about deploying models on real hardware, and, on a personal level, my first extended experience living and traveling in the United States. This report reflects on what I built, what I learned, and how the experience has shaped my thinking about both my research direction and my professional development.
.webp)
.webp)
