Hanxiang Qin (秦瀚翔)

I am a Master’s student in Computer Science at Johns Hopkins University, advised by Alexander Martin and Prof. Benjamin Van Durme. My research focuses on efficient multimodal understanding and retrieval, with a particular emphasis on representation learning and context management.

My recent work scales powerful but costly late-interaction models to be efficient in terms of compute and storage, unlocking their potential for the retrieval of notoriously token-heavy multimodal content spanning text, visual, and audio. We use just a 2.4%-4.9% footprint to preserve or even beat uncompressed state-of-the-art performance. Currently, I’m excited to start my new project on an agentic framework for high-fidelity video understanding.

Before coming to JHU, I was a Software Engineer at Pinduoduo (Temu) for three years, specializing in iOS and cross-platform development. Driven by a core focus on user experience and user growth, I led extensive performance optimizations and framework refactoring. My work empowered rapid product iteration at a massive scale, reliably serving hundreds of millions of daily active users (DAUs).

Previously, I earned a dual degree in Computer Science & Technology and Horticulture from Zhejiang University, where I was advised by Prof. Ming Cai.