Claude Code + NotebookLM + Obsidian = GOD MODE

Highlights

• A unified research workflow combines Claude Code, NotebookLM, and Obsidian to automate data extraction from sources like YouTube and PDFs in under 30 minutes.

• The 'Skill Creator' tool in Claude Code allows users to bundle multiple sub-skills into a single 'super skill' executed via one command.

• NotebookLM offloads token costs and processing from Claude Code, handling complex tasks like generating 15-minute slide decks or infographics for free.

• Storing research in an Obsidian vault creates a transparent file structure that Claude Code uses to learn personal work styles and formatting preferences.

• Updating the 'claude.md' file periodically with latest conversation data turns the AI into a self-improving personal assistant that mimics human reasoning over time.

Timeline

1. 00:00 - 01:32

   The Unified Research Pipeline Architecture

   The integration addresses the need for a practical, repeatable research method rather than isolated AI interactions. While the specific example uses YouTube, the architecture is designed as a template for any professional information source. This setup transforms Claude Code from a simple chat interface into an 'absolute research monster' by leveraging specialized tools for data ingestion and long-term memory.

   • The workflow integrates Claude Code, NotebookLM, Obsidian, and a custom Skill Creator for practical data synthesis.
   • System setup takes less than 30 minutes and remains flexible enough to swap YouTube data for PDFs or articles.
   • The goal is a symbiotic relationship where each tool compensates for the others' limitations.
2. 01:32 - 04:47

   Mechanics of the Autonomous Research Loop

   This workflow operates as a self-improving loop where the more data analyzed, the better the system understands user preferences. Using Markdown files within an Obsidian vault makes the internal data transparent to the AI, allowing it to find information more efficiently than in a standard database. The process effectively creates a 'brain within a brain' that governs how the assistant talks, thinks, and formats output.

   • Claude Code triggers a specific skill to extract YouTube data and send it to NotebookLM for deep analysis.
   • NotebookLM generates deliverables like podcasts, infographics, or slide decks and passes them back to the terminal.
   • Obsidian records every interaction and analysis result in Markdown files to provide a persistent corpus of knowledge.
   • The 'claude.md' file acts as a set of instructions that evolves as the AI observes the user's specific work style.
3. 04:47 - 08:44

   Technical Configuration and Skill Creation

   Installation focuses on the Skill Creator as the central hub for expanding Claude Code's utility. By using the 'notebooklm-py' library, users bypass the lack of a public NotebookLM API, enabling terminal-based control of Google's research tool. A key feature is the AI's ability to 'self-improve' by generating its own skill code based on external documentation provided by the user.

   • The Skill Creator plugin is installed via the '/plugin' command and enables natural language programming of new functions.
   • NotebookLM integration requires the 'notebooklm-py' GitHub repository and CLI authentication via a browser window.
   • The system can autonomously create its own skills by reading documentation from a GitHub repository link.
   • Users can manage up to 50 distinct sources per notebook, including Drive files, text, and YouTube transcripts.
4. 08:44 - 11:35

   Workflow Execution and Result Analysis

   The execution phase demonstrates a real-world test searching for 'MCP servers' to identify market gaps and trending content. This specific run resulted in a Markdown research file and a visual infographic generated without manual intervention. Because NotebookLM handles the heavy lifting, the user avoids high token costs usually associated with feeding large amounts of transcript data into a primary LLM.

   • A single 'YouTube Pipeline' skill executes the search, analysis, and deliverable creation in one step.
   • Offloading processing to NotebookLM saves user tokens since Google handles the heavy computational text analysis.
   • Complex deliverables like full slide decks take approximately 15 minutes, while infographics are ready in a few minutes.
5. 11:35 - 14:31

   Long-term Knowledge Compounding with Obsidian

   The final stage of the workflow focuses on the 'claude.md' file, which serves as a living configuration for the AI assistant. By instructing Claude to update this file based on recent conversations, the user ensures that future research matches their specific tone and depth requirements. This transforms the tool from a generic AI into a highly specialized personal assistant that understands the user's unique 'vibe coding' or professional stack.

   • Obsidian's graph view and backlinking provide human-readable context for AI-generated research files.
   • The 'claude.md' file is updated using a simple prompt to reflect the latest work style and output preferences.
   • Performance gains from this symbiotic relationship compound significantly over months and years of use.