About us at AGIACC
From systems research.To trustworthy AI.
Founded by Xiaoyang Sun, AGIACC connects secure systems research with efficient AI computing—from hardware to applications.
Meet Founder#
Dr. Xiaoyang Sun is the Founder of AGIACC and an expert in high-performance AI infrastructure, secure execution environments, and compiler optimization. He earned his PhD from the University of Leeds and has conducted advanced research at the University of Leeds and the University of Glasgow.
With extensive industrial experience, including AI infrastructure consulting at Kuaishou Technology and research at Alibaba, Xiaoyang bridges cutting-edge academic innovation with enterprise-scale implementation. He has published over 24 peer-reviewed research papers in top-tier computing venues, including SC, OOPSLA, EuroSys, PLDI, ICDCS, SoCC, Cluster, and IEEE TPDS.
At AGIACC, Xiaoyang leads the development of next-generation infrastructure, leveraging hardware-assisted memory safety and compiler optimizations to deliver secure, efficient, and scalable execution for AI agents and enterprise training/inference workloads.
Selected collaborative research
Scroll horizontally to explore.
Protection and optimization
CapOpt
Capability-aware compiler optimization that accounts for memory bounds, permissions and provenance.
Hardware and software evaluation
Sweet or Sour CHERI
Measuring capability-enabled execution on Arm Morello, including systems workloads and LLM inference.
Efficient AI systems
StrongHold
Overlapping computation and memory offloading to train large models within accelerator memory constraints.
Multi-agent systems
Maestro
Workload-aware scheduling that coordinates memory, placement and priorities across multi-agent workflows.
Agent execution
SpecBox
Anticipating sandbox demand and coordinating prewarming, reuse and artifact transport for agent tools.
Language-runtime assurance
COMFORT
Combining generated programs, specification-informed inputs and differential testing to find JavaScript conformance defects.
Cluster resource management
ROSE
Speculative scheduling that uses idle cluster capacity with threshold control and rescheduling.
Full publication record
More
Explore Xiaoyang's publications and citations on Google Scholar.
Google Scholar
Bringing research into spin-out#
Xiaoyang co-founded ACE3 AI Ltd to translate research from his PhD these at the University of Leeds into AI training software addressing accelerator memory constraints.
His research work and contributions received AI SuperConnector support in 2024. In 2025, Xiaoyang received a HiPEAC Technology Transfer Award for the ACE3 technology transfer.
This experience shapes AGIACC’s approach: build working systems, publish reproducible evidence, and test their value with users.
The conviction behind AGIACC#
An AI assistant can read a document, invoke a tool and propose an action. Its permissions need to remain meaningful across applications, runtimes, memory and devices.
AGIACC’s long-term vision is trustworthiness-native AI across the computing stack: chip design and new devices, system software, and application frameworks, SDKs, platforms and services. We focus on system security and execution safety, evaluating protection alongside efficiency.
Build the next stage with us
Help build trustworthy AI infrastructure.
Talk with Xiaoyang about seed investment, a deployment challenge or a research collaboration.