Hao Zhang
Research Scientist at ByteDance
Making Intelligence Cumulative.
My current research focuses on self-evolving agentic systems, particularly agent harnesses and shared context infrastructure.
My broader research spans AI-powered data systems, scalable data and retrieval systems, and hardware-accelerated computing. I have designed and built production-scale systems at ByteDance and Huawei, including trillion-scale vector search and graph databases managing hundreds of billions of edges, with benchmark-leading results.
Research Directions
Agentic Systems & Context Infrastructure
Self-evolving agent systems, evolving harnesses and shared environments, agent memory, and context infrastructure.
AI-Powered Data Systems
LLM-powered semantic queries, AI-integrated data processing, and intelligent query execution.
Scalable Data & Retrieval Systems
Distributed databases, vector search, graph databases, and large-scale query processing.
Hardware-Accelerated Computing
Efficient and portable data processing across GPUs and heterogeneous accelerators.
Current Focus
AutoIA @ ByteDance — Self-Evolving Agentic Systems
I’m developing AutoIA, an IR-first framework for agents to autonomously construct context systems and continuously improve their harnesses and shared environments through experience.
Selected Systems
-
SEMA — LLM-Powered Databases
Brings LLM reasoning into databases, enabling SQL to analyze structured data alongside natural-language content. Achieves 2–10× speedups on evaluated semantic-query workloads.
-
GES @ Huawei — Large-Scale Graph Database
A production graph database supporting hundreds of billions of edges and high-performance, real-time graph queries. Achieved audited #1 LDBC SNB Interactive results on both tracks.
-
TQEX @ Huawei — Hardware-Portable Database Acceleration
Accelerates SQL analytics across NVIDIA and AMD GPUs and Apple chips through portable tensor execution. Achieves 12.2× speedup over DuckDB and 22.7× over HeavyDB on supported TPC-H SF100 queries.
Research Highlights
30+ publications in database systems and AI, including SIGMOD, VLDB, ICDE, COLM, EMNLP, ATC, TKDE, and The VLDB Journal.
#1 on both tracks of LDBC SNB Interactive — Declarative, 2024 and Imperative, 2025 — including a reported 3,000× throughput advantage over the previous #2 on the declarative track.
I received my Ph.D. from the Chinese University of Hong Kong, advised by Prof. Jeffrey Xu Yu and Prof. Hong Cheng, and my B.S. in Computer Science from Wuhan University (Hongyi Honor College · 弘毅学堂).
Selected Honors & Awards
- ByteDance Soaring Star Talent Program（筋斗云人才计划）
- Huawei Gold Award for Individual Excellence — Top 1% (个人金牌)
- Huawei TopMinds Talent Program — First Class (Highest Tier · 天才少年计划)
Connect & Internships
I’m always happy to exchange ideas, discuss research, and explore collaborations on AI and data systems. Feel free to reach out!
For research internships at ByteDance (Shenzhen), please contact me at zhanghaowuda12 [at] gmail [dot] com (email subject: [Intern]).