Building a Personal Knowledge Base — The LearnAI Workflow Guide

A practical, end-to-end guide for building and maintaining a personal knowledge base using Claude Code, Obsidian, and a Jekyll wiki. This is how the LearnAI Doc wiki (96+ entries) was built — and how you can build your own.

Architecture: Three Layers + Four Operations

┌──────────────────────────────────────────────────────────────┐
│                    Knowledge Base System                       │
│                                                               │
│  ┌───────────────┐   ┌────────────────┐   ┌──────────────┐  │
│  │ RAW SOURCES   │   │ WIKI (compiled)│   │ OBSIDIAN     │  │
│  │               │   │                │   │ (personal)   │  │
│  │ Screenshots   │──▶│ _wiki/*.md     │──▶│ Notes        │  │
│  │ URLs / Links  │   │ wiki-index.md  │   │ Quick ref    │  │
│  │ Papers / PDFs │   │ Cover images   │   │ Connections  │  │
│  │ Social posts  │   │ Cross-links    │   │ Action items │  │
│  └───────────────┘   └────────────────┘   └──────────────┘  │
│                              │                                │
│  ┌───────────────────────────▼──────────────────────────┐    │
│  │               SKILLS (automation)                     │    │
│  │                                                       │    │
│  │ /mywiki    → Ingest: source → wiki + obsidian note   │    │
│  │ /query-kb  → Query: synthesize across all entries     │    │
│  │ /lint-kb   → Maintain: health checks, gaps, stale     │    │
│  │ /feynman   → Research: deep dives, lit reviews        │    │
│  └───────────────────────────────────────────────────────┘    │
└──────────────────────────────────────────────────────────────┘

The Three Layers

The Four Core Operations

1. INGEST — Adding New Knowledge

When: You find something interesting (screenshot, URL, paper, tool)

You find something
       │
       ▼
  /mywiki <input>
       │
       ▼
  Detect → Research → Draft → (you review) → Cover → Cross-link → Push → Obsidian note

Skill: /mywiki

/mywiki @screenshot.PNG
/mywiki https://github.com/interesting-repo
/mywiki "I learned about X" https://link1 https://link2
/mywiki ./folder-of-research-notes

The skill handles everything: reading the input, web research, drafting the entry, generating a cover image, cross-linking to related entries, committing/pushing, and creating an Obsidian note. You only review the draft.

2. QUERY — Synthesizing Across Entries

When: You want answers that span multiple entries

/query-kb <question>
       │
       ▼
  Load wiki-index.md → Find relevant entries → Read them → Synthesize with citations

Skill: /query-kb

/query-kb what are all the agent design patterns I've collected?
/query-kb compare all academic research tools in my wiki
/query-kb what connections exist between my AI education and tools entries?
/query-kb what Karpathy insights do I have across all entries?
/query-kb summarize everything added in the last 2 weeks

This is the biggest upgrade over a traditional wiki — you can ask complex questions and get synthesized answers with citations across all 96+ entries, not just browse one at a time.

3. LINT — Health Checking Your Knowledge

When: Monthly, or when the wiki feels stale

/lint-kb [mode]
       │
       ▼
  Check: stale content, missing cross-links, contradictions, gaps, orphans

Skill: /lint-kb

/lint-kb                  # Full health check
/lint-kb stale            # Outdated star counts, versions, claims
/lint-kb gaps             # Topics that should be covered but aren't
/lint-kb cross-links      # Entries that should be linked but aren't
/lint-kb contradictions   # Where two entries say different things

4. RESEARCH — Going Deep on a Topic

When: You want depth, not just a screenshot capture

Skills: Feynman suite

/feynman:deep-research <topic>       # Multi-source investigation
/feynman:literature-review <topic>   # Academic paper survey
/feynman:peer-review <artifact>      # Critique your work
/feynman:paper-code-audit <paper>    # Verify paper claims vs code
/feynman:alpha-research <topic>      # Search arXiv papers
/feynman:watch <topic>               # Track a field over time

Research outputs → feed into /mywiki for wiki entries.

Real-World Use Cases

Scenario A: “I Saw Something Interesting on Social Media”

The most common workflow. Takes ~5 minutes.

1. Screenshot the post → save to your source folder
2. Run /mywiki @screenshot.PNG
3. Claude reads, researches, drafts an entry
4. You review the draft → say “go”
5. Cover image → cross-link → commit → push → Obsidian note — all automatic

Scenario B: “What Do I Already Know About X?”

You’ve been collecting entries for weeks. Time to synthesize.

1. Run /query-kb what do I know about <topic>?
2. Get a cross-entry synthesis with citations to specific wiki entries
3. Use the synthesis for teaching prep, advisor meetings, or identifying gaps

Scenario C: “I Want to Build Deep Knowledge on a Research Topic”

Multi-day research workflow for a topic like formal verification:

Scenario D: “Preparing for a Meeting or Class”

Quick summaries from your existing knowledge.

/query-kb summarize the 5 most important insights from my recent entries
/query-kb what connections exist between harness engineering and AI education?
/query-kb list all tools I've documented with their key features

Scenario E: “Monthly Maintenance”

Keep the knowledge base healthy.

1. /lint-kb → Full report
2. Fix stale entries (outdated star counts, dead links)
3. Add missing cross-links
4. Ask Claude to regenerate wiki-index.md
5. Verify with /lint-kb cross-links

Scenario F: “Sharing Knowledge with the Team”

Your wiki is public — share direct links.

1. /query-kb <their question> → find relevant entries
2. Share: https://weihaoqu.github.io/learnAIDoc/wiki/<category>/<slug>/
3. If not covered: /mywiki "topic" → create it

Skill Reference Card

The Key File: wiki-index.md

This is the master catalog that makes /query-kb work. It lists every wiki entry with title, date, category, tags, and file path — so the LLM can navigate the entire knowledge base from a single file.

Regenerate it after each batch of new entries by asking Claude to scan all _wiki/*.md frontmatter and rebuild the index.

Golden Rules

1. Everything goes through /mywiki — Don’t manually create wiki entries
2. Query before you write — Check if a topic is already covered before creating a new entry
3. Lint monthly — Knowledge rots. Star counts change, tools get renamed
4. wiki-index.md is THE map — Regenerate after every batch of new entries
5. Obsidian = personal, Wiki = public — Keep action items and research connections in Obsidian
6. Use Feynman for depth, /mywiki for breadth — Different tools for different goals
7. Merge when possible — Don’t create a new entry when merging into an existing one is better

Quick Start: Build Your Own

# 1. Set up the wiki repo
mkdir -p my-wiki/_wiki my-wiki/assets/images
cd my-wiki && git init

# 2. Install skills
# Copy query-kb.md and lint-kb.md to ~/.claude/commands/
# Install Feynman: curl -fsSL https://feynman.is/install-skills | bash

# 3. Create your first entry
claude
> /mywiki "I want to document what I learned about <topic>"

# 4. Query your knowledge
> /query-kb what are the key themes in my wiki?

# 5. Health check
> /lint-kb

How LearnAI Team Could Use This

• Use /mywiki to turn screenshots, links, and research notes into reviewed public wiki entries.
• Use /query-kb before meetings, classes, or content planning to synthesize what the team already knows.
• Use /lint-kb monthly to catch stale links, weak cross-links, contradictions, and missing coverage.

Links

• Karpathy’s approach: LLM Knowledge Bases

  llm-wiki gist

• LearnAI Wiki: weihaoqu.github.io/learnAIDoc
• Feynman: github.com/getcompanion-ai/feynman
• Obsidian: obsidian.md