Kevs Obsidian Ingestion Engine — Data skill for Claude Code

Kev's Obsidian Ingestion Engine

A generic, config-driven video content ingestion engine. Point it at a Dropbox folder, define your taxonomy, and it will:

1. Download videos from Dropbox (Google Drive support optional)
2. Transcribe them with Whisper (local or OpenAI API)
3. Extract key frames with OpenCV scene detection
4. Analyze the frames with Claude vision
5. Classify each video using your taxonomy via Claude
6. Write rich markdown notes into your Obsidian vault (with thumbnails, key quotes, suggested clips)
7. Index everything in a Postgres database for search and querying
8. Auto-link people, topics, and locations to build a knowledge graph

Why?

Because your stack of unprocessed footage isn't searchable. This engine reads every video, understands what's in it, and turns it into a navigable Obsidian knowledge graph + queryable database — entirely driven by a single config file.

What you get out

For every video:

• A markdown note in your Obsidian vault with:
  • YAML frontmatter (content type, topics, people, location, tone, quality, confidence)
  • Auto-generated title and one-line summary
  • Embedded thumbnail
  • Key quotes pulled from the transcript
  • Suggested clips with timestamps
  • The full timestamped transcript
  • Wiki-links to people you defined in your taxonomy
• A database row with all the same metadata
• A searchable transcript stored as both full text and timed segments
• Per-frame analysis data (faces detected, sharpness, brightness)
• People appearance tracking for cross-video search

Setup

1. Install

git clone https://github.com/kevinbadi/kevs-obsidian-ingestion-engine.git
cd kevs-obsidian-ingestion-engine
npm install
pip3 install -r requirements.txt

You also need `ffmpeg` and `ffprobe` installed:

brew install ffmpeg     # macOS
sudo apt install ffmpeg # Linux

2. Define your taxonomy

Copy the example and customize it for your project