Hao Zhang

I am an AI-native data systems researcher and builder focused on agentic workflows, retrieval, and accelerator-native execution. At ByteDance, I develop data infrastructure and orchestration frameworks for LLM-based agents. More broadly, my research spans agentic systems, vector and graph databases, and hardware-accelerated query processing for SQL and other data-intensive workloads, with an emphasis on building efficient foundations for modern AI applications.

I received my Ph.D. from the Chinese University of Hong Kong in 2022 under the supervision of Prof. Jeffrey Xu Yu and Prof. Hong Cheng. Before that, I earned my B.S. in Computer Science from the Hongyi Honor School at Wuhan University in 2017.

I have authored and co-authored 20+ papers in top-tier database venues including SIGMOD, VLDB, ICDE, and TKDE. Beyond publications, I build systems: I have designed and architected multiple industry and research platforms, and my graph database work has achieved world-record benchmark results multiple times.

Collaboration and Internships

• I welcome collaboration on technically ambitious systems challenges in database infrastructure, retrieval, vector search, graph systems, and multimodal analytics. Remote collaboration is also welcome.
• We are seeking strong interns in Shenzhen who are excited about research and development in Data+AI systems. If this fits your interests and background, please apply by email and prefix the subject line with [Intern].

Current Research

• Agentic System for LLM orchestration, including [SEMA, Arxiv’26].
• Vector Database + Graph Database for agents, including [SQLVec, ICDE’26],[RapidStore, VLDB’25], [GES, SIGMOD’25], [Graph Storage Benchmark, SIGMOD’25], [Aquila, VLDB’26].
• Hardware-Accelerated Query Engine for structured (e.g. SQL) and unstructured data (e.g., Graph), including [TQEX(SQL), SIGMOD’26], [TenGraph, VLDB’24], and [TGraph, SIGMOD’25].

Prior Research

1. Distributed Query Engine: [Secco, SIGMOD’22], [DISC, VLDB’20], [Crystal, VLDB’18]
2. Graph Algorithms and GNN: [Graph Classification Attack, DASFAA’25], [Time-Dependent Reachability, ICDE’24], [Streaming Embedding, ICDE’24], [Triangle Listing, BigData’16], [Triangle Listing, DPD’17], [Local Clustering Coefficient, DASFAA’17]
3. AI4DB: [ALSS, SIGMOD’21], [Learned Sketch for Subgraph Counting, VLDBJ’23], [NNGP-Card, SIGMOD’22], [Learned Multiway Join, ICDE’21]

Highlights

• Authored and co-authored 20+ papers in top-tier database venues, with recent work spanning vector search, tensor-centric execution, graph systems, and Data+AI infrastructure.
• Designed and architected systems such as SEMA, TDB, GES, SeccoSQL, DISC, and Crystal; the projects page summarizes these systems from research prototypes to production infrastructure.
• Achieved LDBC SNB Interactive world-record results in both the declarative track (2024) and the imperative track (2025), two of the strongest public signals for graph database performance.