Haoqing Wang
Undergraduate Student
Beijing University of Posts and Telecommunications
Research Interests
About
I am an undergraduate student at Beijing University of Posts and Telecommunications (BUPT), expected to graduate in 2028.
I am currently a research intern at Microsoft Research Asia (MSRA), Intelligent Multimedia Group, mentored by Xiaoyi Zhang. I was previously a research intern at Tsinghua University Knowledge Engineering Group (THUKEG), advised by Zijun Yao and Ji Qi.
In open source, I am an Apache Software Foundation Committer and an initial committer and PPMC member of Apache Maka (Incubating), where I work on automated tasks and Computer Use capabilities.
My research interests center on autonomous agents: using computers more effectively, completing long-horizon tasks with greater autonomy, and improving through experience, including a better understanding of users, their preferences, and their workspaces. I currently study Computer Use Agents and (Co)Work Agents, with the goal of making agent capabilities useful to more people, including those who do not code.
I explore these questions through research and open-source product development, drawing inspiration from real-world use. Read more in My Research Interests and Current Thinking.
Selected Publications
View All →Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model
Senqiao Yang*, Kaichen Zhang*, Zhaoyang Jia*, Jinghao Guo*, Yifei Shen*†, Xinjie Zhang*†, Xiaoyi Zhang, Haoqing Wang, et al.
arXiv preprint arXiv:2607.24904
An efficient codec-native streaming multimodal foundation model that reduces visual token consumption by over 75% and achieves up to 3.5x wall-clock inference speedup.
Learning to Respond: A Large-Scale Benchmark and Progressive Learning Framework for Trigger-Centric Online Video Understanding
Jiawen Qian*, Hang Du*, Guoshun Nan, Shan Huang, Jiaqi Yu, Haoqing Wang, et al.
Under review
A large-scale benchmark (TV-Online) and progressive learning framework for trigger-centric online video understanding.
News
Published The Kind of Agent I Want to Build, sharing my research interests in Computer Use, Long-Horizon Agents, and Self-Improvement, along with lessons from building Work Agents and thoughts on their future.
Apache Maka entered the Apache Incubator. I now serve as an Apache Software Foundation Committer and as an initial committer and PPMC member of Apache Maka (Incubating).
Co-authored Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model, now available on arXiv. [project] [code]
Joined the project now known as Apache Maka (Incubating) to explore and build better and more efficient Work Agents, taking responsibility for automated tasks and Computer Use capabilities.
Gave an "AI for Researcher" talk at the School of Electronic Engineering, Beijing University of Posts and Telecommunications (BUPT), sharing the history and current state of coding agents, along with practical usage patterns and tips. [slides]
As project lead, developed Open CoDesign, an open-source version of Claude Design with agentic design capabilities. Designed and built a dedicated agent loop and harness for design workflows from scratch. The project reached 2k stars within 3 days of launch.
Invited by the MSRA Accelerator team to give a talk on "Is AI Reliable for Research?", sharing insights on using coding agents for research and daily workflows with fellow interns and researchers. [slides]
Invited by weishu to join as a collaborator on hapi, working together to build a better hapi.
Honored to be invited by the Qingke Community to join the roundtable at the OpenClaw Vibe Coding & Agent Meetup alongside Qian Yuchao RC, Cheng Jiale, Gaodaiheng Sam, and Fu Jie, for a discussion on Vibe Coding and next-generation agent architectures.
Joined the Intelligent Multimedia Group at Microsoft Research Asia (MSRA) as a research intern, working on Computer Use Agent (CUA) and GUI Agent.
The open-source project OpenCoworkAI/open-cowork was reposted by the leading Chinese AI media account 机器之心.
Joined the OpenCoworkAI/open-cowork team to help develop the open-source version of cowork! This is my first time contributing to an open-source project!
Joined THUKEG, focusing on long-video understanding, token compression, and video + RL related techniques.
Awarded the National Scholarship (top 0.2%).
Co-authored my first paper (under review).
Joined a research lab and began exploring research.
Started my undergraduate studies at Beijing University of Posts and Telecommunications (BUPT).