Research

• Agent Systems

  Temporal long-term memory, identity and data boundaries, planning, tool use, MCP, Skills, evaluation, and human oversight for consequential actions.

• AI Models & Computer Vision

  Architecture, multimodal learning, recognition, segmentation, denoising, inference, and evaluation, with an emphasis on model capability in real tasks.

• Reinforcement Learning & Decision Intelligence

  Distributed environments, reward design, policy learning, game balance, and evaluation methods that reflect real decision quality.

• Production AI

  Retrieval, permissions, observability, deployment, cost, safety, regression testing, and the operating mechanisms that keep systems reliable over time.

• FastAPI-Template · 500+ stars

  A production backend foundation for modularity, RBAC, caching, tasks, observability, and deployment—and the largest project in the portfolio. Production notes ZH

• create-fastapi-app · 100+ stars

  A generator that turns architecture conventions into repeatable service foundations for teams.

• ZipAgent · 100+ stars

  A lightweight Python agent framework covering tool use, streaming, memory, and MCP.

• MiniCode · 70+ stars

  A compact Go coding agent for understanding tool calls, file operations, permissions, and context management. Design notes ZH

• WebSeg · 50+ stars

  A web interface that makes visual segmentation directly usable and interactive.

• Enefit — Predict Energy Behavior of Prosumers · Bronze Medal
• Optiver — Trading at the Close · Bronze Medal
• Child Mind Institute — Detect Sleep States · Bronze Medal

• Author of Enterprise AgentOS Architecture Design and Engineering Practice, published by Tsinghua University Press.
• 15 research publications across AI, computer vision, decision systems, and manufacturing automation.
• 12 authorized invention patents and 12 utility model patents. Clickable public records are in Patents below.
• Nearly 1K stars across public GitHub projects, led by FastAPI-Template with 500+ stars.
• Three bronze medals in international Kaggle machine-learning competitions; the specific competitions are listed above.
• Code and improvement contributions to prominent open-source AI projects including OpenMMLab and sherpa-onnx.
• Enterprise mentor, technical reviewer, and participant in professional computing and robotics communities.

AI, models & algorithms

• CN116912827B

  Granted · Interactive annotation method and system based on large models

• CN119088998A

  Published · Multimodal image retrieval method and system

• CN118734966A

  Published · Real-time inference for heterogeneous image fusion based on deep learning

• CN117058528A

  Published · AI image annotation and training management method and apparatus

• CN116704240A

  Published · Image annotation and training method and apparatus

• CN114266923B

  Granted · Multi-label classification model construction and waste-bottle classification

• CN112766481B

  Granted · Neural-network training and image anomaly detection

• CN111476780B

  Granted · Image detection method, apparatus, electronic device, and storage medium

• CN111310613B

  Granted · Image detection method, apparatus, and computer-readable storage medium

• CN111617478B

  Granted · Game-lineup strength prediction method and apparatus

• CN114254260B

  Granted · Mining unbalanced data groups in games

• CN113230650B

  Granted · Model- and genetic-algorithm-based lineup data processing

• CN115957511A

  Published · Determining virtual-object behavior with trained behavior models

• CN119048478A

  Published · Image clarity evaluation method and apparatus

• CN112384946A

  Published · Neural-network-based defective-pixel detection

Software & data systems

• CN116701505A

  Published · Intelligent conversion of waste-sorting data

• CN116468429A

  Published · Renewable-waste value estimation method and system

• CN119311937A

  Published · Remote collection of device information

• CN110457214A

  Published · Application testing method, apparatus, and electronic device