An Obsidian knowledge graph + Claude Code agent + Supabase backbone that captures every apprentice, every employer partner, every event, every sponsor — and a voice AI named Linna who talks about it. Built the way you asked for it.
Andrej Karpathy's "AI reads, summarizes, extracts relationships, builds the graph autonomously" — except instead of a personal journal, it's the entire operational map of Dakar Foundation.
Your vault has six hub notes — Dakar Pathway, Cinema Basel West, Employer Partners, Sponsors, Academic Partners, Compliance — and one file per apprentice, employer, sponsor, event. One concept, one file.
Every note [[wiki-links]] to related notes. Andre Robinson's apprentice file links to [[Lighting & Grip]], [[LA City College]], and [[Dakar Pathway]]. Obsidian renders the relationships in Graph View.
Drop emails, PDFs, or notes into 99-Inbox/. A Python harness running Claude Opus 4.7 reads each one, extracts entities (people, partners, events), writes atomic notes, and wikilinks them automatically.
The same data lives in a Supabase Postgres — 11 tables for tracks, apprentices, employers, sponsors, equipment, events, placements, sponsor-track sponsorships, compliance items. Wikilinks become foreign keys.
The Smart Connections plugin generates embeddings on every note for semantic search. Dataview turns the vault into a queryable database — "show me every active apprentice in the AI in Media track" returns a live table.
A Vapi voice agent (Claude Haiku + ElevenLabs) trained on the entire vault. Students, employers, sponsors, and journalists can phone in and Linna routes them. Same knowledge, two surfaces.
This is your vault's live graph. Drag a node. Hover for the excerpt. Watch clusters form around hubs — apprentices around tracks, tracks around the Dakar Pathway, partners around sponsors.
Obsidian's lightest convention — numbered top-level folders so the index reads itself.
Click a note to see its raw markdown — that's what the agent reads, that's what Smart Connections embeds, that's what Dataview queries.
Pick a note on the left to view its raw markdown →
Drop a CSV of new sponsor leads, a PDF of a partnership agreement, or a forwarded email into 99-Inbox/. The rest happens by itself.
Your data lives in plain markdown + plain Postgres. Anything we build on top is replaceable. You're not renting your knowledge.
Obsidian vault — plain markdown files on your filesystem. Wikilinks, YAML frontmatter, atomic notes. Open it in any text editor, sync it to iCloud, version it in Git.
Claude Code + Claude Opus 4.7 running locally. Reads the inbox, extracts entities, writes notes, wikilinks them, archives sources. One Python harness. Configurable in plain markdown.
Supabase Postgres — 11 tables modeling the entire Dakar operation. Public read on academic partners + events. Service-role for staff CRUD. Foreign keys mirror the wikilinks.
Linna — a Vapi-powered voice agent with Grace's voice clone. Routes students, employers, sponsors, journalists. Available 24/7 from any phone or browser.
Your existing dakarpathway.lovable.app stays in place. This tour deploys alongside it. Plug into your CRM, your email, your social — without re-architecting anything.
DOL + CA DAS audit-ready exports out of Supabase. Apprentice rosters, hours logged, employer placements, certifications — queryable in plain SQL when the auditor calls.
The vault, the agent, the Supabase, the voice layer — all wired and running. Next move is a 30-minute call to walk you through running it locally, dropping your first real email into the inbox, and watching it become structured knowledge.