6.9 KiB
6.9 KiB
llm-wiki-sync: Turning Scattered Notes into a Reusable Knowledge Base
Overview
A cyclical pipeline showing how raw notes are continuously transformed through LLM-powered ingestion into structured wiki pages, then feedback into the knowledge base for reuse.
Learning Objectives
The viewer will understand:
- The continuous circular flow of llm-wiki-sync from raw notes to reusable knowledge
- The six key extraction outputs: Summary, Claims, Quotes, Entities, Concepts, Connections
- How feedback and reuse complete the cycle back to new raw material
Section 1: The Circular Flow (Center Concept)
Key Concept: The llm-wiki-sync pipeline operates as a continuous cycle, not a linear process.
Content:
- 7 stages in the cycle: Raw Note → Ingest → Extract → Source Page → Graph/Site → Reuse → Feedback Loop
- Each stage feeds into the next, with feedback returning to the beginning
- The cycle is continuous and self-reinforcing
Visual Element:
- Type: circular flow diagram
- Subject: 7 stages arranged in a circle with clockwise arrows
- Center label: "llm-wiki-sync Cycle"
- Treatment: chalk style with hand-drawn arrows connecting stages
Text Labels:
- Headline: "The Knowledge Pipeline Cycle"
- Stage labels: "Raw Note", "Ingest", "Extract", "Source Page", "Graph/Site", "Reuse", "Feedback"
- Center: "llm-wiki-sync"
Section 2: Stage 1 — Raw Note (Input)
Key Concept: Raw notes are the starting point of the cycle.
Content:
- Original documents stored in raw/ folder
- Can be any format: markdown, text, research notes
- Contains unprocessed information awaiting structure
Visual Element:
- Type: document/note icon
- Subject: Stack of paper or note icon
- Treatment: Chalk sketch style
Text Labels:
- Label: "Raw Note"
- Description: "Original input"
Section 3: Stage 2 — Ingest (LLM Analysis)
Key Concept: The LLM analyzes raw notes and extracts structured information.
Content:
- Hermes skill triggers Claude Code for ingestion
- LLM reads and analyzes the full document
- Context check against wiki/index.md prevents duplicates
Visual Element:
- Type: brain/processing icon
- Subject: Brain or gears with chalk lines
- Treatment: Hand-drawn chalk illustration
Text Labels:
- Label: "Ingest"
- Description: "LLM Analysis"
Section 4: Stage 3 — Extract (Six Core Outputs)
Key Concept: Six key elements are extracted from each document.
Content:
- Summary: 核心主题, 问题域, 方法/机制, 结论/价值
- Key Claims: Verifiable assertions extracted from text
- Key Quotes: Preserved citations for reference
- Key Entities: Named people, companies, products (e.g., LaunchDarkly, HP)
- Key Concepts: Abstract terms that can be reused (e.g., RTO, RPO, Feature Flag)
- Connections: Relationships between elements (depends_on, enables, provides)
Visual Element:
- Type: 6 callout nodes around center
- Subject: Six boxes or bubbles representing extraction outputs
- Treatment: Chalk circles with icons inside each
Text Labels:
- Headline: "Extraction Outputs"
- Labels: "Summary", "Claims", "Quotes", "Entities", "Concepts", "Connections"
Section 5: Stage 4 — Source Page (Structured Output)
Key Concept: Extracted information is written as a structured wiki source page.
Content:
- Written to wiki/sources/.md
- Contains frontmatter (id, title, type, tags, sources, last_updated)
- Standard sections: Summary, Key Claims, Key Quotes, Key Concepts, Key Entities, Connections, Contradictions
- Links use PageName format for interconnections
Visual Element:
- Type: document/page icon
- Subject: Page with visible structure headers
- Treatment: Chalk sketch with text lines
Text Labels:
- Label: "Source Page"
- Description: "wiki/sources/*.md"
Section 6: Stage 5 — Graph & Static Site
Key Concept: Structured pages generate knowledge graphs and static websites.
Content:
- graph.json: Machine-readable graph structure
- graph.html: Interactive visualization
- Quartz: Static site generation for sharing/export
- Connections become edges in the knowledge graph
Visual Element:
- Type: network/graph icon
- Subject: Connected nodes representing knowledge graph
- Treatment: Chalk diagram with nodes and edges
Text Labels:
- Label: "Graph & Site"
- Description: "graph.json + Quartz"
Section 7: Stage 6 — Reuse (Knowledge Application)
Key Concept: Structured knowledge enables multiple reuse scenarios.
Content:
- Query: Ask questions against the knowledge base
- Synthesize: Create new content from existing knowledge
- Write: Generate articles, reports from source material
- Connect: Link ideas across different source pages
Visual Element:
- Type: multiple arrows pointing outward
- Subject: Reuse scenarios as icons (question, document, pen)
- Treatment: Chalk illustration
Text Labels:
- Label: "Reuse"
- Sub-labels: "Query", "Synthesize", "Write", "Connect"
Section 8: Stage 7 — Feedback Loop (Continuous Cycle)
Key Concept: Reuse generates new raw notes, completing the cycle.
Content:
- New insights from synthesis become new raw notes
- Updated knowledge feeds back to raw/ folder
- Cycle continues indefinitely
- Each iteration strengthens the knowledge base
Visual Element:
- Type: circular arrow
- Subject: Feedback loop arrow returning to Raw Note stage
- Treatment: Large chalk arrow completing the circle
Text Labels:
- Label: "Feedback Loop"
- Description: "New notes → Cycle repeats"
Data Points (Verbatim)
Key Quotes
- "RTO is about speed: how fast you get back online. RPO is about data: how much you can afford to lose."
- "Deploy != Release. Feature flags change this. You can deploy code to production without releasing it to users."
Key Entities
- LaunchDarkly (Feature Flag management platform)
- HP (example enterprise)
- Christian Dior (example case)
Key Concepts
- RTO (Recovery Time Objective)
- RPO (Recovery Point Objective)
- Feature Flag (特性开关)
- Kill Switch (紧急关闭机制)
- 渐进式发布 (Gradual Rollout)
Design Instructions
Layout Preferences
- Circular flow with 7 stages evenly spaced around a circle
- Clockwise arrow direction
- Center contains the main concept "llm-wiki-sync"
- Each stage is a node with icon + label
- Extraction outputs (6 items) shown as callouts or inner ring
Style Preferences
- Chalkboard: Dark background (#1A1A1A)
- Hand-drawn chalk style for all elements
- Chalk colors: white, yellow, pink, blue, green, orange
- Imperfect, sketchy lines throughout
- Chalk dust effects for authenticity
Text Requirements
- All text in English
- Legible font sizes (minimum 14pt for labels)
- Clear hierarchy: title > headlines > labels > descriptions
- Ample whitespace between stages
Visual Clarity
- Avoid clutter - only essential elements
- Each stage should be clearly distinguishable
- Arrows should clearly indicate flow direction
- Publication-ready quality