Research & Engineering
Jiayu Xu
Agent Systems · AI Models · Production Engineering
Focused on agent systems, AI models, and production engineering that connect to real data, tools, and workflows.
Previously worked on AI systems at Tencent IEG, led software and algorithms at Guangdong DataBeyond Technology, and later led a 50+ person R&D organization delivering enterprise AI platforms.
Experience
From algorithm research to AI platforms and R&D leadership.
- Technical path
- 10+ years across computer vision, reinforcement learning, model engineering, and agent platforms.
- Organization & delivery
- Led a 50+ person R&D organization across technical direction, architecture, quality, and delivery.
- Education & research
- B.Eng., M.Eng., and Ph.D. from Harbin Institute of Technology; postdoctoral research at SIAT, CAS, jointly trained with Tencent.
- Public record
- A published AgentOS monograph, 15 papers, 24 authorized patents, nearly 1K GitHub stars, and three bronze medals in international machine-learning competitions.
Focus areas
Long-term work across agents, models, and production systems.
-
Enterprise Agent Systems
Planning, memory, tools, Skills, and MCP with permissions, audit, and evaluation.
-
Model, Data & Evaluation Loops
Connected workflows for data, annotation, training, evaluation, deployment, and feedback.
-
Production AI Platforms
Latency, cost, observability, security, and release quality.
-
AI R&D & Teams
Connecting technical judgment, team execution, and delivery.
Selected work
Agent platforms, industrial multimodal AI, model systems, and real-time AI.
-
Enterprise Agent Platform
Models, knowledge, tools, permissions, and evaluation across 12 systems at 99.93% availability.
-
Industrial Multimodal & Engineering Agents
Drawing and 3D model understanding, similar-part retrieval, process generation, and quotation support.
-
Model & Data Loop
120+ datasets, 2M+ images, and 100+ models; accuracy improved from 92% to 99.5%.
-
EvoAI Vision Model Workbench
A private workflow for data, annotation, training, testing, and inference delivery.
-
Reinforcement Learning & Real-time AI
Distributed training on 1,500+ CPU cores and anomaly detection across 300 devices.
-
Longitudinal Personal AI Platform
Long-term context, controlled tool use, and human confirmation for key actions.
Book
An engineering guide to enterprise agent architecture and delivery.

Enterprise AgentOS Architecture Design and Engineering Practice
Architecture, runtimes, MCP, RAG, permissions, security, evaluation, and platform operations.
Selected research
Five selected papers from a record of 15 publications.
Writing
Notes on models, agents, and production engineering.
-
Models
-
Agents
-
Engineering
For AI teams, agent platforms, or production deployment, contact xujiayu0606@gmail.com.