Hongbin Zhong(钟宏斌)CS PhD Student
Georgia Institute of Technology |
![]() |
I am a second-year PhD student in Computer Science at Georgia Tech, fortunate to be advised by the kind and supportive Prof. Kexin Rong.
I also closely collaborate with Dr. Adriana Szekeres from Microsoft Research and Dr. Nina Narodytska from VMware Research, as well as Dr. Matthew Lentz from Duke University.
Previously, I was a research intern at Microsoft Research Redmond during the summer of first year. Before my PhD, I was also an intern at Columbia University, working with Prof. Eugene Wu, where I contributed to the publication of multiple DB top-tier conference papers.
Additionally, I have extensive industry research/engineering experience, including but not limited to LLM agents and planning, core database systems, and server development, which you can learn more about on my LinkedIn.
My research interest lies in AI agents, where I focus on the entire stack, from the underlying computer systems to the algorithmic and application layers. I am deeply interested in both low-level systems and pure AI research. Coming from the systems community, I firmly believe that true growth comes from stepping beyond one’s comfort zone. Therefore, I do not want to confine myself solely to systems research; I aspire to explore all layers of the AI stack, so that I can develop a more comprehensive perspective and capability to solve complex problems.
Key themes I have led as the first and only student author, which equipped me with rich hands-on experience:
Beyond Screenshots: An Dynamic State-Machine Memory and Global Programmatic Planner for WebAgents
Hongbin Zhong* with Microsoft Research People
HoneyBee: Efficient Role-based Access Control for Vector Databases via Dynamic Partitioning
Hongbin Zhong*, Matthew Lentz, Nina Narodytska, Adriana Szekeres, Kexin Rong
arxiv 2025(Under Revision in SIGMOD 2026)
[paper]
Fast Hypothetical Updates Evaluation
Haneen Mohammed*, Alexander Yao*, Charlie Summers*, Hongbin Zhong, Gromit Yeuk-Yin Chan, Sub- rata Mitra, Lampros Flokas, Eugene Wu
SIGMOD 2025 DEMO
FaDE: More Than a Million What-ifs Per Second
Haneen Mohammed*, Alexander Yao*, Charlie Summers*, Hongbin Zhong, Gromit Yeuk-Yin Chan, Sub- rata Mitra, Lampros Flokas, Eugene Wu
VLDB 2025
[code]
Accelerating Deletion Interventions on OLAP Workload
Haneen Mohammed, Alexander Yao, Lampros Flokas,Hongbin Zhong, Charlie Summers, Eugene Wu
ICDE 2024
PECJ: Stream Window Join on Disorder Data Streams with Proactive Error Compensation
Xianzhi Zeng*, Shuhao Zhang, Hongbin Zhong, Hao Zhang, Mian Lu, Zhao Zheng, Yuqiang Chen
SIGMOD 2024
Microsoft Research May. 2025 - Aug. 2025, Redmond, Seattle, US closely worked with Adriana Szekeres, Suman Nath Focus: Architected the Beyond Screenshots WebAgent planner/memory stack, raising WebArena success to ~90%. |
|
Georgia Institute of Technology Aug. 2024 - Present, Atlanta, US Advisor: Kexin Rong; Collaboration with VMware Systems Group Focus: Built RBAC-aware vector database partitioning (13.5x faster, 90% less memory) and joint batching for RAG pipelines. |
|
InfiniFlow Apr. 2024 - Jun. 2024, Shanghai, China Vector Databases Contributor Focus: Reworked timestamp persistence and bulk-deletion cleanup to cut vector storage latency and I/O. |
|
4Paradigm Feb. 2024 - Jun. 2024 Part-time Full-stack Engineer Intern Focus: Tuned AI assistant caching for lower latency and shipped async community features with scheduled refresh. |
|
Columbia University Jul. 2023 - Dec. 2023, New York, US closely worked with Eugene Wu Focus: Optimized FADE sparse-matrix evaluation with SIMD/multithreading for near-linear 8x speedups. |
|
Rutgers University Jun. 2023 - Sep. 2023 closely worked with Dong Deng Focus: Implemented similarity-search baselines and parallel group-function analytics pipelines. |
|
Nanyang Technological University Jan. 2023 - Jul. 2023 Focus: Built low-latency disorder stream processing and Bayesian variational inference for complex event data. |
|
Meituan Apr. 2022 - Sep. 2022, Beijing, China Backend Engineer Intern Focus: Delivered short-video backend features, Kafka/Hive recommendation pipelines, and periodic refresh for low-bandwidth users. |