Turn any YouTube Channel into your AI Mentor - Obsidian is the ultimate automation workbench for PKM

TL;DR

Obsidian automation can fetch YouTube transcripts and convert them into structured Markdown notes with summaries, key takeaways, and timestamped transcript links.

Briefing Cornell Notes

Briefing

A practical way to turn any YouTube channel into an “AI mentor” for Obsidian is taking shape: feed Google’s Notebook LM with condensed, timestamped transcripts and summaries, then ask questions and get answers that point back to the exact video moments. The payoff is simple—when a user has a specific how-to question (like enabling a particular Excalidraw panel or custom tools), the system can retrieve the relevant guidance from hundreds of videos and dozens of release notes without manually hunting through playlists or search results.

The core bottleneck is twofold. First, Notebook LM can only ingest up to 50 “resources,” but a typical creator’s library can easily exceed that—here, 250 YouTube videos plus 400 detailed release notes. Second, Notebook LM needs the text in a form it can read, which requires obtaining YouTube transcripts. The solution is a pipeline built around Obsidian automation: two scripts (created with help from ChatGPT) fetch video data and generate structured Markdown “notes” for each video. Each output includes a summary, key takeaways, and the full timestamped transcript, with clickable links that jump directly to the corresponding moment in the YouTube video.

To make the dataset small enough for Notebook LM, the transcripts are bundled into quarterly files—so the system can process a manageable number of resources while still preserving the ability to trace answers back to specific videos. In a live example, a question about turning on the “custom tools panel” (including “custom pens”) is answered by routing the user to the exact Obsidian settings path and the relevant video segment. Another example addresses a misunderstanding about assigning properties to Excalidraw drawings, steering the user toward the correct document-property approach and pointing to the specific documentation file and video guidance.

Once the Notebook LM workspace is created from these quarterly bundles, users can ask questions against the condensed knowledge base and receive targeted responses. The creator also demonstrates that Google does not allow public sharing of Notebook LM workspaces, so the workaround is to distribute the training files. Users can extract a provided folder, create a Notebook LM account, and drop the files in—reportedly taking less than a minute.

The approach scales beyond one channel. With the same Obsidian scripts, users can download transcripts from other YouTube channels, generate the Markdown training materials, and then load them into Notebook LM. A demo uses Nicole van der Hovind’s channel: the downloader pulls recent videos, produces quarterly bundles, and those bundles can then be queried in Notebook LM about that channel’s content.

Underpinning everything is the claim that Obsidian is the automation workbench for PKM: APIs, templating, data view, Excalidraw integration, and scripting turn note-taking into a system for collecting, transforming, and interrogating knowledge. The announcement ends with a community launch—Technov Visual PKM Community—positioned as a place to learn and troubleshoot this automation workflow, including scripting basics and how to use AI tools alongside Obsidian.

Cornell Notes

The workflow turns a large YouTube library into an “AI mentor” by feeding Notebook LM with condensed, timestamped transcripts and summaries. Obsidian automation solves two constraints: transcripts must be fetched and Notebook LM’s 50-resource limit must be respected by bundling videos into quarterly Markdown files. Each generated note includes key takeaways plus the full transcript with clickable timestamps, so answers can be traced back to exact moments in the source videos. After loading the quarterly bundles into Notebook LM, users can ask detailed questions (e.g., how to enable Excalidraw-related panels or clarify property misconceptions) and get targeted guidance. The same pipeline can be reused for other channels by downloading their transcripts and regenerating the Notebook LM training materials.

Why does the project need both transcript fetching and transcript condensation before Notebook LM can help?

Notebook LM can only ingest up to 50 resources, but the knowledge base can exceed that quickly (250 YouTube videos plus 400 release notes). That forces condensation—here, videos are bundled into quarterly files. At the same time, Notebook LM needs readable text, so transcripts must be obtained from YouTube. The Obsidian scripts handle both: one fetches video data, and another generates Markdown notes that include summaries, key takeaways, and timestamped transcript text.

How do the generated notes preserve “where to look” inside long videos?

Each video’s Markdown output includes a timestamped transcript where paragraphs link back to the corresponding YouTube time. That means when Notebook LM answers a question, the user can click through to the exact segment in the original video rather than relying on a generic summary.

What does “quarterly bundling” accomplish in the Notebook LM workflow?

Quarterly bundling reduces the number of Notebook LM inputs by grouping multiple per-video notes into fewer larger files. Instead of trying to load hundreds of individual transcripts, the system loads a smaller set of quarterly bundles, staying within Notebook LM’s resource cap while still retaining traceability via timestamps inside each bundle.

How does the system handle questions that are really about Obsidian/Excalidraw settings rather than general knowledge?

The pipeline indexes the creator’s own documentation content—video guidance plus release-note material—so Notebook LM can retrieve the correct settings path and the relevant segment. In the examples, questions about enabling the “custom tools panel” and “custom pens” are answered with the correct Obsidian settings route and linked to the specific video section that demonstrates it.

How can someone use the same approach for a different YouTube channel?

The downloader vault and Obsidian scripts can fetch transcripts for any channel. The demo swaps in Nicole van der Hovind’s channel URL, downloads recent videos, generates the Markdown training files (including quarterly bundles), and then those bundles are loaded into Notebook LM for question answering about that channel’s content.

Why can’t the Notebook LM workspace simply be shared publicly, and what replaces it?

Google doesn’t allow public sharing of Notebook LM workspaces. Instead, the workflow distributes the training files so users can create their own Notebook LM account, drop the files into a new workspace, and start asking questions locally.

Review Questions

What two constraints (one about input size, one about input format) determine the design of the pipeline, and how does the workflow address each?
How does quarterly bundling help stay within Notebook LM limits while still allowing users to navigate to exact video timestamps?
If you wanted to adapt this system to a new YouTube channel, what steps would you repeat in Obsidian before asking questions in Notebook LM?

Key Points

1
Obsidian automation can fetch YouTube transcripts and convert them into structured Markdown notes with summaries, key takeaways, and timestamped transcript links.
2
Notebook LM’s 50-resource limit is handled by bundling many per-video notes into quarterly files rather than loading each video separately.
3
Clickable timestamps inside the generated notes let answers be traced back to exact moments in the source videos.
4
The same pipeline can be reused for other channels by downloading their transcripts and regenerating the quarterly training bundles.
5
Notebook LM workspaces can’t be publicly shared, so distributing training files is the practical workaround.
6
Obsidian is positioned as the central “automation workbench” by combining scripts, templating, data organization, and AI-assisted processing.
7
A community launch is planned to support learning and troubleshooting of this PKM automation workflow.

Highlights

The system turns hundreds of videos into a queryable knowledge base by condensing them into quarterly Notebook LM bundles.

Generated transcripts include clickable timestamps, so answers can jump directly to the relevant video segment.

The workflow is portable: swap in a different channel URL, regenerate the training files, and ask Notebook LM questions about that new channel’s content.

Notebook LM sharing is blocked publicly, but distributing the training files enables anyone to recreate the workspace quickly.

Topics

AI Mentor
Notebook LM
Obsidian Automation
YouTube Transcripts
PKM Workbench