ArXiv Second Brain

A personal research wiki built on Karpathy's LLM Wiki pattern, currently focused on market microstructure. Synthesised by Claude Opus 4.7 from arXiv papers, published as a static site.

Live site: see the Pages tab once deployed.

Structure

Three layers:

• raw/papers/ — arXiv PDFs + full-text markdown conversions (immutable source).
• wiki/ — LLM-generated synthesis pages organised by type (papers, concepts, methods, entities, comparisons, connections, mindmaps).
• CLAUDE.md + .claude/skills/ — schema and operation protocols that drive how the wiki is built.

See the site Home page for the full layer-architecture explanation and wiki-vs-RAG comparison.

Local development

Install dependencies:

pip install -r requirements.txt
pip install mkdocs-material pymdown-extensions mkdocs-mermaid2-plugin

Preview locally:

python scripts/sync_wiki_to_mkdocs.py
cd site && mkdocs serve

Opens at http://127.0.0.1:8000.

Ingesting a new paper

See .claude/skills/arxiv-ingest/SKILL.md for the full procedure. Short form:

python scripts/ingest_arxiv.py <arxiv_id>
# Then open Claude Code and say: "ingest the new paper in raw/"

Deploy

Every push to main triggers .github/workflows/deploy-pages.yml which builds and publishes to GitHub Pages.

License

Source material (PDFs in raw/papers/) retains its original arXiv licensing. Wiki synthesis content is owned by the repo owner.