HRyu
← Notes

AI & Technology

Second Brain

The methodology behind this wiki system. Sources: Karpathy's LLM Wiki pattern, Tiago Forte's PARA/CODE, Hakyun's AI Second Brain build guide

Second Brain

The methodology behind this wiki system. Sources: Karpathy's LLM Wiki pattern, Tiago Forte's PARA/CODE, Hakyun's AI Second Brain build guide.

The Core Idea (Karpathy)

Most LLM-document systems are RAG: retrieve chunks at query time, regenerate from scratch every time. Nothing accumulates.

The wiki pattern is different: the LLM incrementally builds and maintains a persistent artifact. When a new source arrives, it's read, synthesized, and integrated — updating entity pages, flagging contradictions, strengthening existing synthesis. Knowledge compounds.

"The wiki is a persistent, compounding artifact. The cross-references are already there. The contradictions have already been flagged."

Human role: Source curation, exploration, asking good questions. LLM role: Summarizing, cross-referencing, filing, bookkeeping.

Three-Layer Architecture

Layer What It Is Who Owns It
Raw sources Articles, papers, images, data files — immutable source of truth Human
Wiki LLM-generated markdown: summaries, entity pages, concept pages, synthesis LLM
Schema (CLAUDE.md) Conventions, workflows, how to ingest/query/maintain Co-evolved

This vault's schema is at CLAUDE.md/CLAUDE.md.md.

Operations

Ingest

Drop a source into raw-sources/, tell the LLM to process it.

Flow: Read source → discuss key takeaways → write summary page → update index → update related entity/concept pages → append to log.

A single source typically touches 5–15 wiki pages.

Query

Ask questions against the wiki. LLM reads index first, drills into relevant pages, synthesizes with citations.

Output forms: markdown page, comparison table, Marp slide deck, matplotlib chart, canvas.

Key insight: Good answers can be filed back as new wiki pages — explorations compound just like sources.

Lint (Health Check)

Periodic check for:

  • Contradictions between pages
  • Stale claims superseded by newer sources
  • Orphan pages (no inbound links)
  • Important concepts mentioned but lacking their own page
  • Missing cross-references
  • Data gaps that could be filled with web search

Index + Log Structure

index.md — Content-oriented catalog. Every page listed with link + one-line summary + optional metadata. LLM reads this first on every query to find relevant pages.

log.md — Chronological, append-only. Format: ## [YYYY-MM-DD] ingest | Source Title. Parseable with grep "^## \[" log.md | tail -5.

Tiago Forte's PARA + CODE

PARA (Organization)

  • Projects — Active work with a defined end
  • Areas — Ongoing responsibilities (health, finances, relationships)
  • Resources — Reference material
  • Archive — Inactive items

CODE (Workflow)

  • Capture — Get ideas out of your head
  • Organize — File into PARA structure
  • Distill — Extract key insights (progressive summarization)
  • Express — Produce outputs (writing, projects, presentations)

Key principles

  • Search is the primary retrieval mechanism (not perfect filing)
  • Second brain improves output production by offloading cognitive load
  • Build for future self who needs to find it quickly

Hakyun's Philosophy for This System

From raw-sources/Life and Finances/1. SECONDBRAIN - AI pipeline.md:

  • Obsidian as local markdown knowledge graph
  • AI as thinking partner, not passive storage
  • Feedback loop: refine → accept/reject AI links → improve
  • Avoid passive dump — focus on atomic notes + aggressive linking + AI as synthesis partner

"A system that thinks with me, not just stores information."

Tooling Notes

  • Obsidian Web Clipper — browser extension, converts articles to markdown
  • Obsidian graph view — best way to see connection topology, hub pages, orphans
  • Dataview plugin — runs queries over YAML frontmatter (tags, dates, source counts)
  • Marp — markdown slide deck format; useful for generating presentations from wiki content
  • qmd — local hybrid BM25/vector search for markdown at scale

Karpathy's LLM Wiki — Full Pattern Details

Source: raw-sources/AI Projects/Karpathy's LLM wiki.md (Karpathy, 2026)

This is the canonical source for the pattern this vault is built on. Key additions beyond what was already summarized:

Operations in full:

  • Ingest: Read source → discuss key takeaways → write summary page → update index → update 5–15 related pages → append log. "Personally I prefer to ingest sources one at a time and stay involved."
  • Query outputs: markdown page, comparison table, Marp slide deck, matplotlib chart, canvas. Good answers should be filed back as new pages — explorations compound.
  • Lint: Contradictions, stale claims, orphan pages, missing cross-references, data gaps.

Why it works: "The tedious part of maintaining a knowledge base is not the reading or the thinking — it's the bookkeeping. Humans abandon wikis because the maintenance burden grows faster than the value. LLMs don't get bored."

Vannevar Bush connection: Related in spirit to the Memex (1945) — personal curated knowledge store with associative trails. Bush's vision was private, actively curated, with connections as valuable as documents. The part he couldn't solve: who does the maintenance. LLMs handle that.

Optional CLI: qmd — local hybrid BM25/vector search for markdown files (when wiki grows past index-file scale).

How I Built My Second Brain — Obsidian Approach

Source: raw-sources/AI Projects/How I Built My Second Brain with Obsidian.md (Kevin T'Syen, 2025)

Alternative Vault Structure (for reference)

00. Inbox        → scratchpad; dump and sort later
01. Daily Notes  → log, tasks, journal (one note/day)
02. Projects     → each project: meeting notes, todos, planning, references
03. Code Snippets → reusable functions, terminal commands
04. Architecture & Design → system diagrams, technical decisions
05. Learnings    → summaries of books, courses, articles

Why Obsidian Over Notion

Factor Obsidian Notion
Data ownership Local markdown (yours) Closed ecosystem (theirs)
Offline Fully offline Needs internet
Speed Fast even at scale Sluggish with large databases
Version control Git (full history, free) Basic undo only
Customization Hundreds of plugins Limited integrations
Dev-friendly Markdown native; query with Dataview Block-based, not text-native

Free sync via Git: Obsidian Git plugin → auto-commits to private GitHub repo. Free syncing + full version history.

Key plugins:

  • Excalidraw — Diagrams embedded directly in notes
  • Kanban — Task management inside Obsidian
  • Templater — Consistent note structure (daily log, meeting notes, bug reports)
  • Tasks — Recurring tasks, filters, due dates
  • Dataview — SQL/JS queries over note frontmatter (surface open tasks, recent notes)

Monetising the Second Brain

Source: raw-sources/AI Projects/Monetising Secondbrain.md

The second brain is infrastructure, not the product. It becomes valuable when it saves time, makes money, or improves decisions for someone else.

Six Monetisation Paths

1. Done-for-You Second Brain Systems (fastest to monetise) Build custom Obsidian + AI workflows for clients:

  • Target: financial advisors, founders, analysts, content creators
  • Sell: decision systems, workflow automation, time saved
  • Pricing: $300–$2,000 per setup

2. AI Knowledge Operator Service Combine Obsidian (storage) + Claude/GPT (reasoning) + system design. Becomes a personal AI analyst for the client. "The guy who builds AI brains for professionals."

3. Niche Product (scalable) "Second Brain for Watch Collectors" — Hakyun's unfair edge:

  • Collection tracker, strap pairing database, market price tracking, brand wiki
  • Sell as Notion/Obsidian template ($19–$79) or premium AI-feature version

4. Analytics + Second Brain Hybrid Merge data skills with knowledge system → "decision intelligence system." For a business: store data insights, link sales trends + staffing decisions + forecasts, then layer AI to answer "why did revenue drop?"

5. Internal Tool for Companies AI-powered company wiki, internal knowledge systems, decision dashboards. Solves: scattered docs, lost knowledge, inefficient workflows.

6. Content → Audience → Monetization Second brain as content engine and idea generator. Publish insights/frameworks on LinkedIn, Substack. Monetize via consulting, templates, paid content.

The 3 Most Realistic Paths for Hakyun

  1. High-ticket service — Build systems for finance people, analysts, founders
  2. Watch niche product — No one else has his level of niche interest + systems thinking
  3. AI + analytics hybrid — "I build decision systems for businesses" (extends Merlion Café-type analytics work)

⚠️ Trap to avoid: "I need to perfect the system first." Don't. Need 70% working system + 30% real-world feedback.

Related Pages

[[active-projects]] | [[hakyun-ryu]] | [[index]] | [[claude-code-tools]]