Agentic AI
从 Chatbot 到生产级 Agent：这个合集，我想认真讲清楚 AI Agent 开发
A roadmap for learning agent systems from message handling and model calls through tools, RAG, orchestration and production boundaries.
Technical writing
Making complex ideas easier to learn.
A selection of Chinese-language articles that turn implementation experience into practical explanations of algorithms, AI systems and engineering choices.
Content topics
5Reinforcement Learning · Machine Learning · Deep Learning · Agentic AI · EngineeringPublication activity
06Reinforcement learning · machine learning · agents · engineeringLanguage
ZHChinese01 / Featured articles
Agentic AI
Agent 开发教程（1）：为什么你的第一个 Agent 项目，应该从 Chatbot 开始
Explains why message contracts, streaming, persistence and failure handling form the foundation beneath an agent interface.
Reinforcement Learning
动态规划教程 02：二维动态规划，用最小路径和理解网格状态
Uses minimum path sum to make two-dimensional dynamic-programming state, transition and boundary conditions concrete.
02 / All topics
01Agentic AI
从 Chatbot 到生产级 Agent：这个合集，我想认真讲清楚 AI Agent 开发
A roadmap for learning agent systems from message handling and model calls through tools, RAG, orchestration and production boundaries.
02Agentic AI
Agent 开发教程（1）：为什么你的第一个 Agent 项目，应该从 Chatbot 开始
Explains why message contracts, streaming, persistence and failure handling form the foundation beneath an agent interface.
03Reinforcement Learning
动态规划教程 02：二维动态规划，用最小路径和理解网格状态
Uses minimum path sum to make two-dimensional dynamic-programming state, transition and boundary conditions concrete.
04Machine Learning
面试必备：PCA 主成分分析算法原理
A geometric and statistical explanation of PCA, dimensionality reduction and the distinction from nonlinear visualization methods.
05Deep Learning
弱监督学习是什么？从 Kaggle RSNA 膝关节 MRI 竞赛看真实世界为什么没有完美标签
Connects weak supervision to clinical imaging, imperfect labels and the evaluation boundaries of a real competition workflow.
06Engineering
LLM、知识管理与 Vault Coach
A project-facing explanation of why source-grounded answers, practice and governed knowledge structures belong in one learning workflow.