AI Knowledge Library with Historical Personas
Build a personal knowledge library where AI personas (Turing, Curie, and more) write editorials from your notes. A Zettelkasten-powered learning system using Claude Code.
You take hundreds of notes from books, podcasts, and articles — but they just sit there. You never synthesize them into anything useful, and the connections between ideas stay locked in your head.
Who it's for: lifelong learners with growing note collections, Obsidian and Zettelkasten users, students doing independent research, writers building thematic bodies of work, curious generalists
Example
"Write an editorial about consciousness from my neuroscience notes" → Alan Turing persona synthesizes your 15 atomic notes into a 1,500-word editorial with citations back to your sources, then Marie Curie writes a response challenging his conclusions
New here? 3-minute setup guide → | Already set up? Copy the template below.
# Editorial Library
## Goal
Generate insightful editorials by synthesizing atomic notes through configurable personas. Each editorial should surface unexpected connections between ideas and present them in the persona's unique voice.
## Directory Structure
- `notes/` — Atomic notes with YAML frontmatter (tags, source, date)
- `personas/` — Persona configuration files (one .md per persona)
- `editorials/` — Generated editorials (YYYY-MM-DD_persona_topic-slug.md)
- `responses/` — Persona responses to other editorials
- `tags/` — Auto-generated tag index files
## Note Format
Each note in `notes/` should follow this structure:
```
---
title: "Note Title"
source: "YouTube video / Article / Book / etc."
sourceUrl: "https://..."
tags: ["tag1", "tag2", "tag3"]
date: "YYYY-MM-DD"
questions: ["Question this raises?", "What connects to?"]
---
Key insight or atomic idea from the source. One idea per note.
```
## Persona Format
Each persona in `personas/` defines a voice:
```
---
name: "Ada Lovelace"
era: "Victorian England, 1840s"
expertise: ["mathematics", "computation", "philosophy"]
style: "Analytical yet poetic, draws connections between art and science"
influences: ["Charles Babbage", "Mary Somerville", "Lord Byron"]
---
Write as Ada Lovelace would — blending mathematical precision with creative imagination.
Reference the analytical engine and computational thinking where relevant.
Draw unexpected parallels between mechanical processes and artistic expression.
```
## Editorial Generation Rules
1. Select 3-5 notes with overlapping or complementary tags
2. Choose a persona whose expertise aligns with the note cluster
3. Generate an editorial that synthesizes the notes into a cohesive narrative
4. The editorial must reference specific notes (by filename) as sources
5. End with a concluding perspective, not open-ended questions
6. Auto-generate new tags from the editorial content
7. Save to `editorials/` with format: YYYY-MM-DD_persona_topic-slug.md
## Response Generation Rules
1. Select an existing editorial
2. Choose a different persona to respond
3. The response should engage with specific arguments from the original
4. Maintain the responding persona's authentic voice and perspective
5. Save to `responses/` with format: YYYY-MM-DD_persona_responds_original-slug.md
## Commands
- "Create a new note from [source]" — Extract atomic notes from a URL, text, or file
- "Generate an editorial" — Find related notes and generate a new editorial
- "Have [persona] respond to [editorial]" — Generate a persona response
- "Show me note clusters" — Find groups of notes with overlapping tags
- "Update the tag index" — Rebuild the tag index files
- "What hasn't been covered?" — Find notes not yet used in any editorial
What This Does
This playbook creates a personalized editorial library where AI personas (based on historical figures or custom characters) generate essays and editorials by synthesizing your curated atomic notes. It's like having an on-demand team of tutors who write about topics you find interesting, in styles that make learning feel natural. Inspired by a Reddit user who built an entire editorial website with personas like Turing, Curie, Borges, and even a Neanderthal responding to each other's work.
Prerequisites
- Claude Code installed and configured
- A dedicated project folder for your knowledge library
- Source material: notes, bookmarks, YouTube summaries, article highlights, or any content you want to synthesize title: "Note Title" source: "YouTube video / Article / Book / etc." sourceUrl: "https://..." tags: ["tag1", "tag2", "tag3"] date: "YYYY-MM-DD" questions: ["Question this raises?", "What connects to?"]
Key insight or atomic idea from the source. One idea per note.
## Persona Format
Each persona in `personas/` defines a voice:
name: "Ada Lovelace" era: "Victorian England, 1840s" expertise: ["mathematics", "computation", "philosophy"] style: "Analytical yet poetic, draws connections between art and science" influences: ["Charles Babbage", "Mary Somerville", "Lord Byron"]
Write as Ada Lovelace would — blending mathematical precision with creative imagination. Reference the analytical engine and computational thinking where relevant. Draw unexpected parallels between mechanical processes and artistic expression.
## Editorial Generation Rules
1. Select 3-5 notes with overlapping or complementary tags
2. Choose a persona whose expertise aligns with the note cluster
3. Generate an editorial that synthesizes the notes into a cohesive narrative
4. The editorial must reference specific notes (by filename) as sources
5. End with a concluding perspective, not open-ended questions
6. Auto-generate new tags from the editorial content
7. Save to `editorials/` with format: YYYY-MM-DD_persona_topic-slug.md
## Response Generation Rules
1. Select an existing editorial
2. Choose a different persona to respond
3. The response should engage with specific arguments from the original
4. Maintain the responding persona's authentic voice and perspective
5. Save to `responses/` with format: YYYY-MM-DD_persona_responds_original-slug.md
## Commands
- "Create a new note from [source]" — Extract atomic notes from a URL, text, or file
- "Generate an editorial" — Find related notes and generate a new editorial
- "Have [persona] respond to [editorial]" — Generate a persona response
- "Show me note clusters" — Find groups of notes with overlapping tags
- "Update the tag index" — Rebuild the tag index files
- "What hasn't been covered?" — Find notes not yet used in any editorial
Step-by-Step Setup
Step 1: Create the project structure
mkdir -p ~/editorial-library/{notes,personas,editorials,responses,tags}
cd ~/editorial-library
Step 2: Create your first personas
Create personas/turing.md:
---
name: "Alan Turing"
era: "Mid-20th century England"
expertise: ["computation", "mathematics", "artificial intelligence", "cryptography"]
style: "Precise and logical, but with dry wit and thought experiments"
influences: ["Church", "Godel", "Russell"]
---
Write as Turing would — with mathematical rigor and a fascination for what machines can and cannot do. Use thought experiments to illustrate points.
Create at least 2-3 personas with different expertise areas for interesting cross-pollination.
Step 3: Add your first notes
Create notes from content you've consumed. One idea per file:
notes/distributed-consensus-raft.md:
---
title: "Raft Consensus Algorithm"
source: "MIT 6.824 Lecture"
tags: ["distributed-systems", "consensus", "fault-tolerance"]
date: "2026-01-30"
questions: ["How does this relate to human decision-making?"]
---
Raft achieves consensus by electing a leader who manages log replication. If the leader fails, a new election occurs. The key insight is that consensus requires a majority, not unanimity.
Step 4: Save the CLAUDE.md and launch
Save the template above as CLAUDE.md, then:
cd ~/editorial-library
claude
Try: "Generate an editorial from my notes on distributed systems"
Example Usage
Create notes from a YouTube video:
"Create atomic notes from this transcript: [paste transcript]. Extract 3-5 key ideas, each as a separate note file with appropriate tags."
Generate an editorial:
"Look at my notes tagged with 'emergence' and 'complexity'. Have Ada Lovelace write an editorial synthesizing these ideas."
Create a response chain:
"Have the Neanderthal persona respond to Turing's latest editorial on computation. Focus on embodied knowledge vs abstract reasoning."
Discover connections:
"Show me note clusters — which groups of notes share the most tags but haven't been combined into an editorial yet?"
Build a reading session:
"Generate 3 short editorials from my unprocessed notes, each by a different persona. Focus on notes I added this week."
Tips
- Curate first, generate second: The quality of editorials depends entirely on the quality of your notes. Spend time curating good atomic notes before generating.
- Diverse personas create better content: A Neanderthal responding to a mathematician creates more interesting perspectives than two similar personas.
- Use tags as gravity: Tags pull related ideas together. The more consistent your tagging, the better the note clustering.
- Let personas respond to each other: The most interesting content emerges from multi-turn persona dialogues, not standalone editorials.
- One idea per note: Keep notes atomic. A single clear insight is more useful than a long summary.
Troubleshooting
Problem: Editorials feel generic or "AI-like"
Solution: Make your persona files more specific. Include writing quirks, preferred metaphors, intellectual blind spots, and strong opinions. The more constrained the persona, the more distinctive the voice.
Problem: Notes aren't clustering well
Solution: Review your tags for consistency. Use a controlled vocabulary — don't have both "AI" and "artificial-intelligence" as separate tags. Ask Claude to "normalize my tags across all notes."
Problem: Editorials don't reference source notes
Solution: Remind Claude in your prompt to cite specific note filenames. The CLAUDE.md rules require this, but explicit prompting helps reinforce it.