Knowledge Management

Knowledge Management is a two-phase personal system: automatic ingestion of what I read and watch, and public visualization as an interactive concept graph. The core idea — when I write a post or article, the research cards are already ready. No more searching during writing.

Why this project

Before, the creation process looked like this: find an interesting concept in a podcast → note it somewhere → forget it → find it again → search for sources → finally write.

The problem isn’t access to information. It’s the friction between “I heard something interesting” and “I have a structured, sourced card ready to publish.”

This project removes that friction in two layers:

1. Ingestion: send a link on Telegram, receive a .md card ready for Ekenor
2. Visualization: published cards form a navigable graph — like Obsidian, but in the browser

Phase 2 — Concept Graph (delivered ✅)

The visualization is live at /concepts/map. A 3D force graph with 5 thematic clusters, 17 nodes, and links between related concepts.

Architecture:

• _graph.json: single source of truth — contains slug, title, summary (≤15 words), cluster, related concepts
• ConceptMap.tsx: React component — WebGL rendering via 3d-force-graph (not react-force-graph which bundles A-Frame)
• map.astro: Astro page with client:only="react" (WebGL canvas incompatible with SSR)
• ConceptSearch.tsx: client-side Fuse.js search + cluster filters

Interactions:

• Hover → tooltip with cluster + summary + hint
• Click → opens the concept card in a new tab
• “Recenter” button → zoomToFit(400, 80)
• Nodes in the same cluster attract each other — inter-cluster links create visual “bridges”

Updates: /sync-concepts command — Claude reads the existing index (slugs + summaries), reads only new cards, decides cluster + links, updates _graph.json without re-reading the existing ones.

Phase 1 — Knowledge Ingestion Pipeline (in progress)

Send a Telegram link (YouTube, article, PDF) → the workflow fetches the content → GPT extracts key concepts with sources → generates .md files ready for src/content/concepts/.

Planned stack:

• n8n for orchestration
• yt-dlp + Whisper for videos
• Scraping + GPT for web articles
• Automatic generation of Ekenor frontmatter (slug, cluster, relatedConcepts)

What changes

Without this system: read → note → forget → search → write.

With this system: read → send a link → card ready → publish.

The quality of articles and posts improves mechanically — sources are there, structured, at the moment of writing. Not during.

Phase 2 delivered in March 2026. Phase 1 under construction.