AI 时代的知识库工程
Universal Knowledge Base Engineering
从研究中心到领域知识库 · 学会用 AI 编译的方式构建可持续生长的知识体系
Knowledge Base Engineering for the AI Era
Universal Knowledge Base Engineering
From research lab to domain knowledge bases · Build living knowledge systems the AI way
市面上的 Obsidian / Notion 教程只教你"怎么写",UKBE 教你"怎么让 AI 编译你的知识"。四大差异,决定了这门课的独特价值。
Most Obsidian / Notion tutorials teach you how to write. UKBE teaches you how to let AI compile your knowledge. Four differentiators define its unique value.
不是 RAG 碎片检索。AI 像资深编辑一样把原始资料重写为结构化知识页面,把推理压力前置到存储前。
Not fragmented RAG retrieval. AI acts as a senior editor — it rewrites raw material into structured knowledge pages, shifting reasoning cost upstream of storage.
以 File-over-App、Compilation vs RAG、Knowledge Linting、Human-AI Feedback Loop 四大原则为方法论根基,不是网红技巧拼盘。
Built on four principles: File-over-App, Compilation vs RAG, Knowledge Linting, Human-AI Feedback Loop. A coherent methodology — not a bag of influencer hacks.
先学建方法论中台(UKB-Lab),再用中台产出服务多个领域知识库。避免每个项目从零摸索。
First build a methodology lab (UKB-Lab). Then reuse it to serve multiple domain knowledge bases. Stop starting from scratch for every project.
所有规则(如 R-8)都由受控对照实验(EXP 系列)得出,不是主观经验。学员学完也能自己设计实验、升级自己的 SOP。
Every rule (e.g., R-8) comes from controlled experiments (EXP series), not from opinion. Graduates design their own experiments and evolve their own SOPs.
这不是四条孤立技巧,而是一套有因果关系的体系 —— File-over-App 是根,Compilation 是方法,Linting 是保障,Feedback Loop 是演进引擎。
Not four isolated tricks — a causal system. File-over-App is the root, Compilation is the method, Linting is the guardrail, Feedback Loop is the evolution engine.
知识的寿命必须长于处理它的软件。Markdown + Git 是底层,不依附任何具体工具。
Knowledge must outlive the software that processes it. Markdown + Git is the substrate — not tied to any specific app.
AI 不是搜索引擎,是资深编辑。预先把原始资料编译成结构化 Wiki,换来跨文档的逻辑一致性。
AI is an editor, not a search engine. Pre-compile raw material into structured wiki pages — and gain cross-document consistency.
像代码 lint 一样扫知识库:断链、矛盾、冗余、孤岛,AI 主动识别并提建议。
Lint your knowledge base like code — detect dead links, contradictions, redundancy, and orphan pages. AI flags and proposes fixes.
人喂料,AI 编译,图谱反哺人。最终由人通过 Graph View 审视知识密度,发现新研究方向。
Humans ingest raw material. AI compiles. The graph feeds insight back to humans — who spot new directions by reading topology.
M4(搭建研究中心)和 M7(实验驱动)是整门课的价值锚点,其他模块都围绕这两点服务。每个模块都有明确的学习目标和学员产出,落地可验收。
M4 (Build Your Research Lab) and M7 (Experiment-Driven) are the course's value anchors — every other module serves these two. Each module has explicit learning goals and student deliverables.
| # | 模块Module | 学习目标Learning Goal | 学员产出Deliverable |
|---|---|---|---|
| M0 | 导论:为什么要建知识库Why Build a Knowledge Base | 理解 RAG 局限与 Compilation 范式的必然性Grasp RAG's limits and why Compilation is inevitable | 一份个人知识盘点清单A personal knowledge audit |
| M1 | Karpathy 知识工程哲学Karpathy's Philosophy | 掌握四大原则及其内在逻辑Master the four principles and their logic | 用四原则评估一个现有知识库Audit an existing KB against the four principles |
| M2 | 工具栈基础Tool Stack | 独立配齐 Markdown + Git + Obsidian + AI 助手Wire up Markdown + Git + Obsidian + AI assistant | 本地跑通的最小知识库A minimal working KB, locally |
| M3 | 标准作业程序 SOPStandard Operating Procedure | 掌握盘点/提案/缝合/校验四步 + R-1~R-8 红线Inventory / Propose / Stitch / Verify + R-1–R-8 guardrails | 自己项目的 SOP 适配版A project-tailored SOP |
| M4 | 实战:搭建你的研究中心 🔑Build Your Research Lab 🔑 | 把 M1–M3 组合落地Integrate M1–M3 into a working lab | 自己的 UKB-Lab(结构 + git + Obsidian + SOP)Your own UKB-Lab (structure + git + Obsidian + SOP) |
| M5 | 可视化与 Knowledge LintingVisualization & Linting | 掌握 Graph View + Dataview + 死链扫描Graph View + Dataview + dead-link scanning | 自己知识库的拓扑健康报告A topology health report of your KB |
| M6 | 原子化条目设计Atomic Entry Design | YAML 8 字段 + WikiLink 拓扑设计8-field YAML + WikiLink topology design | 5 条高质量原子条目 · 零死链5 high-quality atomic entries · zero dead links |
| M7 | 实验驱动的方法论沉淀 🔑Experiment-Driven Methodology 🔑 | 学会用对照实验升级自己的 SOPEvolve your SOP via controlled experiments | 一份自己的 EXP 报告Your own EXP report |
| M8 | 下游知识库搭建(案例库)Domain KB (Case Library) | 看懂 3 类典型领域的建库模式Understand 3 typical domain patterns | 为自己选定的领域搭建首版知识库Ship a first-version KB for your chosen domain |
| M9 | 进阶:Templater / AI Skills / 自动化Advanced: Templater / AI Skills / Automation | 用自动化降低人工成本Reduce manual overhead with automation | 至少一个自动化流程投入使用At least one automation in production use |
| M10 | 发布与演进Publish & Evolve | Cloudflare Pages 发布 + 持续维护Cloudflare Pages deploy + ongoing maintenance | 可公开访问的知识库A publicly accessible knowledge base |
这门课最大的不同:它不是一次性写完的静态课程。每当你在真实项目中产出一条新规则、一次新实验、一个新案例 —— 课程会主动反哺自己。你学到的永远是最新鲜的一手经验。
The biggest difference: this is not a one-shot static course. Every time you produce a new rule, run a new experiment, or ship a new case in real work — the course feeds itself back. You're always learning the freshest firsthand insight.
为什么这很重要? 静态课程在 AI 时代的半衰期非常短。UKBE 把"课程"和"真实研究"同轴绑定 —— 研究在动,课程就在动。学员学到的不仅是当前版本的知识,更是一套会自我升级的活体系统。
Why this matters. Static courses age quickly in the AI era. UKBE binds "course" and "real research" on the same axle — when research moves, the course moves. You don't inherit a snapshot. You inherit a living system that upgrades itself.
别被名字吓到。UKBE 的核心目标学员不是"已经是知识工程师的人",而是"意识到自己的知识资产到了必须升级管理方式"那一刻的人。
Don't be intimidated by the name. UKBE isn't built for "people already known as knowledge engineers" — it's for anyone at the moment they realize their knowledge assets need a real management upgrade.
回到顶部重新浏览大纲,或直接跳到你最关心的模块开始规划。
Scroll back up to browse the syllabus, or jump straight to the module you care about most.
查看 10 个模块 Browse 10 Modules