research-os

A personal research operating system that turns NotebookLM notebooks into a searchable, linkable, source-grounded Obsidian knowledge vault.

What it is

An opinionated pipeline for academic/industry literature review that glues together:

• NotebookLM (via the notebooklm-py CLI) for paper ingestion, Q&A, and artifact generation (mind maps, briefing docs, study guides, quizzes, flashcards, slide decks)
• Obsidian as the primary reading and linking surface, with:
  • Dataview for auto-generated indices and reading queues
  • Mindmap NextGen for interactive, zoomable concept maps
• Claude Code skills (notebooklm-research, obsidian-research-vault) that codify the end-to-end workflow so new papers are ingested with a single prompt

Directory layout

research-os/
├── obsidian-vault/              ← open this in Obsidian
│   ├── _index.md                    Dataview-driven MOC
│   ├── literature/<topic>/          one sub-folder per topic; one .md per paper
│   ├── maps/<topic>/                mindmap.md (heading-only) + mindmap.json
│   ├── distilled/<topic>/           briefing-doc / study-guide / quiz / flashcards
│   ├── projects/                    flat; cross-cutting research threads
│   └── experiments/                 flat; cross-cutting experiment logs
├── templates/                   ← used by the obsidian-research-vault skill
│   ├── literature-note.md           dossier template (callouts, footnotes, transclusion)
│   ├── project-note.md              with Dataview-queryable frontmatter
│   └── experiment-note.md
├── sources/                     ← raw PDFs downloaded for NotebookLM ingestion
├── exports/                     ← slide decks, infographics, CSVs (consumed outside Obsidian)
├── scratch/                     ← raw probe JSON from NotebookLM Q&A runs (provenance)
└── scripts/                     ← future helpers

Core idea: grounded dossiers, not skeletons

Each paper becomes a 300–600 line literature-note dossier in obsidian-vault/literature/<topic>/. The dossier uses:

• Obsidian callouts (> [!abstract], > [!quote], > [!warning], > [!question], etc.) for visual hierarchy
• Footnote-style source anchors ([^s1] → paper section + NotebookLM source UUID) for every non-trivial claim
• ![[...]] transclusion of distilled artifacts (flashcards, quizzes, study guides) inside collapsible callouts
• Rich frontmatter (rating, status, relevance, confidence, key_claims, topic) so Dataview queries can surface "unread high-relevance papers in topic X"

The Mind2Web 2 paper under obsidian-vault/literature/mind2web2-agentic-search/ is the worked example — compare *.md (current dossier) against *.v1.md (the thin prior template) for the before/after.

Getting started

1. Install Obsidian and open the vault

1. Install Obsidian.
2. Open the obsidian-vault/ directory as an Obsidian vault.
3. Settings → Community plugins → Turn on → install Dataview and Mindmap NextGen (both already listed in .obsidian/community-plugins.json). Trust community plugins on first prompt.

2. Install the NotebookLM CLI (optional — only if you want to regenerate)

The vault content was generated via the notebooklm-py community CLI. Install from its project page, then run notebooklm login to authenticate with a Google account that has NotebookLM access.

3. Hook up the Claude Code skills (optional)

If you use Claude Code, the pipeline is automated via two skills under ~/.claude/skills/:

• notebooklm-research — NotebookLM ingestion, Q&A (12-probe ask --json pattern), and artifact generation
• obsidian-research-vault — vault organization (dossier assembly, Dataview-queryable frontmatter, Markmap-compatible heading-only mindmaps)

Tell Claude Code: "Ingest <arxiv-url> into the <topic-slug> notebook" and it will ingest the paper, run the 12-probe Q&A, generate artifacts, and drop a full dossier into obsidian-vault/literature/<topic>/.

Conventions

• Topic slug — lowercase-hyphenated, short, descriptive. e.g., mind2web2-agentic-search, kv-cache-inference.
• Literature, maps, distilled — nested under the topic slug.
• Projects, experiments — flat, since they tend to span topics.
• Exports — flat, prefixed with the topic slug: exports/mind2web2-slides.pptx.

See the skill files (~/.claude/skills/notebooklm-research/SKILL.md, ~/.claude/skills/obsidian-research-vault/SKILL.md) for the full rules.

License

Personal research workspace. No license granted; do not redistribute paper content.