Summary
Highlights
The speaker introduces an automated system to track, transcribe, AI analyze, and organize YouTube videos into Obsidian. This system eliminates manual tracking and can be applied to any YouTube channel. It transforms each video into a searchable, organized note within Obsidian, complete with transcripts, AI analysis, video statistics, and key insights.
The video demonstrates the structure of an automated note in Obsidian, using 'NANE2 Obsidian in three steps' as an example. Each note includes YAML properties with video information, video stats (views, likes, comments, publish date, duration), a hidden description, quick links, chapter information, a full transcript (generated by a self-hosted Whisper AI Docker container), and AI analysis from Gemini, providing a brief summary, best ideas, tools, resources, and key learning points. Archive details are also included.
Phase one focuses on setting up the workflow and its decision logic. It begins with a manual trigger, fetches YouTube videos via the YouTube API, and reads existing tracking data from a Google Sheet. A 'merge and process' node compares new videos against spreadsheet data. Smart decision logic determines if a video is new (adds with 'to-do' status), existing but needing processing ('to-do' or 'failed'), or already processed ('done'), preventing re-processing and handling failed attempts.
Phase two is where videos are transformed into Obsidian notes. It processes videos one by one to avoid overload, fetches complete YouTube metadata, and formats the data. A Python app downloads audio files, which Whisper AI then converts to text. Google Gemini generates AI analysis from the transcript, extracting key ideas and learning points. Finally, all data sources are merged into a comprehensive package ready for Obsidian, including video stats, transcript, AI analysis, quick links, and archive details.
The final phase involves delivering the processed content and managing the workflow's continuation. Each YouTube video note is saved into the Obsidian vault using a local REST API. The Google Sheet is updated, marking the video as 'done' with an archive date. The system checks if more videos need processing and loops back if so, or ends gracefully if all are complete. This self-managing system ensures continuous, systematic processing and archiving of videos, taking 5-10 minutes per video.
A live demonstration shows the workflow tracking notes in a Google Sheet, initiating the Python YouTube transcriber, and executing the workflow in NADN. The process involves fetching new videos, updating their status to 'to-do' in the spreadsheet, transcribing audio with Whisper AI, analyzing with Gemini, and finally, creating and updating notes in Obsidian. The demo highlights the seamless automatic update of video status in the spreadsheet and the creation of detailed Obsidian notes.
The speaker concludes by outlining future improvements, including enhancing transcript quality using 'get recall' for cleaner outputs and easier timestamp navigation. Another idea is to allow processing of third-party YouTube channels by simply changing the channel ID. A particularly exciting prospect is adding Retrieval Augmented Generation (RAG) to enable conversational search within the Obsidian vault, turning the video library into a searchable knowledge base. Finally, the speaker considers exporting the entire workflow as a JSON for public use, inviting community feedback on these ideas.