AI Native · independent work · public portfolio

方法先于模型,
证据先于结论。

以 AI Native 作为工作方式,把工程判断归纳成可检验的方法。这里整理围绕 Agent、数据平台与工程交付的独立研究和匿名化案例——重点不在技术清单,而在问题如何被界定、边界如何收束、结论如何验证。

从三个入口开始。

项目案例 · anonymized cases

两个主案例

面向专业材料的 RAG 知识构建,与基于静态 SQL 证据的字段级血缘治理——把知识与 SQL 分别收束为可核验、可审计的结论。

进入项目索引

文章 · independent research

两篇文章

Agentic 边界悖论与 Loop Engineering——讨论人机边界如何移动、开发 Agent 如何成为可验收的循环。

阅读文章

简历 · career summary

历年总结

这些判断来自十余年数据与 Agent 工程实践。历年经历、技术栈与项目细节按时间线展开。

查看简历

AI 编程视频

六集工程讲解,围绕并行智能体、主编排、RAG 的噪音、黄金测试集、AGENTS.md 与需求边界。

点击播放后才会向 bilibili.com 发起请求。

工具 · project Q&A agent

项目问答智能体

只基于经公开审查的项目报告回答;证据不足时显式说明边界,不猜测。

直接提问

AI Native · independent work · public portfolio

Methods before models,
evidence before conclusions.

I work AI-native and turn engineering judgment into methods that can be checked. This site collects independent research and anonymized cases around agents, data platforms, and engineering delivery — the point is not a technology list, but how a problem is framed, how its boundary is narrowed, and how a conclusion is verified.

Start from three entrances.

Projects · anonymized cases

Two flagship cases

RAG knowledge building for professional material, and field-level lineage governance grounded in static SQL evidence — each reduced to a verifiable, auditable conclusion.

Open the project index

Articles · independent research

Two articles

The agentic boundary paradox and loop engineering — how the human–machine boundary moves, and how agent development becomes an acceptable loop.

Read the articles

Resume · career summary

A decade of practice

These judgments come from more than ten years of data and agent engineering practice. Roles, stacks, and project details unfold along a timeline.

View the resume

AI coding videos

Six engineering episodes on parallel agents, coordinator orchestration, RAG noise, golden test sets, AGENTS.md, and requirement boundaries.

Requests to bilibili.com are made only after you click play.

Tool · project Q&A agent

Project Q&A agent

Answers only from publicly reviewed project reports; when evidence is insufficient, the boundary is stated explicitly instead of guessing.

Ask it directly