Second Brain – An AI-Operated Knowledge Vault
What It Is
This is the system I actually run my work on, not a demo. It's an Obsidian vault, a single connected body of 898 atomic notes tied together by 6,674 wikilinks, and it doubles as my second brain across every front I operate. What makes it more than a folder of markdown is how it's driven: Claude Code sits on top of the vault as an operator, with automated pipelines that turn meetings, email, research and daily captures into structured, linked notes instead of raw dumps. The video renders the entire graph as a force-directed map, captured from Obsidian's graph view and accelerated 5x, so the growth and the clustering read at a glance. The graphic layer, the telemetry HUD, the domain legend and the animated final count, was authored as code with HyperFrames (video rendered from HTML) and the numbers on screen are pulled live from the vault, not mocked. It's a portrait of a working AI-operated knowledge system, told through the system's own data.
Features
The vault is the substance and the video is the artifact: I built the composition as code (HyperFrames renders video from an HTML/GSAP timeline), sped the screen capture 5x with ffmpeg, embedded the type, and overlaid a telemetry HUD with a domain legend and a live count of 898 notes and 6,674 links measured from the vault at render time. Nothing on screen is a placeholder.
Technical Highlights
Obsidian Knowledge Graph
898 notes connected by 6,674 wikilinks, visualized with Obsidian's force-directed graph view. Nodes are colored by domain (work, content, personal, research) so clusters emerge from how ideas link, not from the folder tree.
Operated with Claude Code
Claude Code runs as the operator on top of the vault, reading and writing notes under a strict protocol. Automated pipelines ingest meetings, email and research and atomize them into linked notes, so the graph grows continuously instead of by hand.
Self-Auditing System
Graph queries, semantic embeddings and nightly health checks keep the map connected, flagging orphan notes, broken links and structural drift so the knowledge base stays navigable as it scales.
HyperFrames Motion Design
The overlay, HUD, kinetic titles and count-up were authored as code with HyperFrames, which renders deterministic video from an HTML and GSAP timeline. The 1080p master was produced entirely from the composition, no timeline-based editor.
Key Features
PROJECT TYPE: Personal AI knowledge infrastructure
SCALE: 898 notes, 6,674 wikilinks
OPERATOR: Claude Code on an Obsidian vault
VISUAL: Force-directed graph, colored by domain
PRODUCTION: HyperFrames (video-from-code) + ffmpeg 5x speed-up
STYLE: Telemetry HUD, kinetic type, live data
Results
Turned an abstract personal system, a second brain, into a concrete, legible visual that shows scale and connectivity in under 35 seconds.
Demonstrated an AI-operated knowledge workflow end to end: agents ingest and link notes, the graph grows, and the system audits its own connectivity.
Produced a broadcast-quality 1080p piece from code alone, using HyperFrames for the motion layer and ffmpeg for the 5x timelapse, with every on-screen number pulled live from the vault.
Built a reusable template: swap the footage, the beats and the domain colors and the same composition becomes a release or product video for other projects.