Skip to content

Repository files navigation

llm-wiki

A personal AI Engineering knowledge base, built on Andrej Karpathy's LLM Wiki pattern.

Rather than re-deriving knowledge on every query, this wiki compiles sources into structured, interlinked markdown pages that compound over time. New sources update existing pages. The graph gets richer with every ingest.

Maintained by Luke Hornof — AI Engineering Leader | founder (acq.) · PhD in CS · CTO · Intel Nervana


How It Works

Three operations, driven by Claude Code + the schema in CLAUDE.md:

Operation What it does
Ingest Drop a source into _raw/ → Claude extracts entities and updates 10–15 wiki pages
Query Ask a question → Claude reads relevant pages and synthesizes an answer with citations
Lint Health check → finds orphan pages, broken links, stale content, contradictions

Sources come primarily from Twitter/X, Reddit, podcasts, and GitHub. The Obsidian Web Clipper sends articles to _raw/ in one click.


Structure

_raw/          # Immutable source drops (gitignored)
sources/       # One page per ingested source
tools/         # AI tools and frameworks
models/        # Model families and providers
concepts/      # Foundational and emerging AI concepts
people/        # Key researchers and practitioners
companies/     # AI labs and startups
topics/        # Themes spanning multiple entities
tutorials/     # Hands-on guides and walkthroughs
index.md       # Content catalog
log.md         # Append-only operation history
CLAUDE.md      # Schema — entity page templates, workflows, conventions

Current Contents

As of May 2026: 19 tools · 34 people · 22 concepts · 25 companies · 3 models · 4 topics · 2 tutorials · 97 sources.

See index.md for the full catalog.

A few load-bearing pages a new reader might start with:


Build Your Own

Start your own LLM Wiki in about 10 minutes.

Prerequisites

Steps

  1. Create a folder with a _raw/ subdirectory
  2. Open Claude Code in that folder
  3. Paste Karpathy's gist and tell Claude to help you set up an LLM Wiki — it will ask what your wiki is about, then write a CLAUDE.md schema tailored to your use case
  4. Drop sources into _raw/ and tell Claude: "ingest new content"
  5. Open the folder as an Obsidian vault for the graph view

For a detailed walkthrough, see OrewaDeveloper's Medium article.


Credits

Based on Andrej Karpathy's LLM Wiki gist. Explained well by OrewaDeveloper's Medium article.

About

AI engineering knowledge base — tracks models, tools, people, and concepts across the LLM landscape. Built and maintained with Claude Code.

Resources

Stars

4 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors