◇ Case study / 05 · 2026
WSJR Knowledge Graph
Internal ops wiki · Obsidian
◇ The problem
What we were trying to solve
Three group entities sharing one operational reality but no shared brain — onboarding docs, vendor contacts, deploy runbooks, AI agent prompts were scattered across Google Docs, Notion pages, WhatsApp screenshots, and people's heads. Anything that hadn't been written down in the last week was effectively lost.
◇ The solution
How it shipped
Built a structured Obsidian vault with a top-level WSJR/ folder containing: per-entity sub-folders (Investments, TopStocx, School), shared infrastructure docs (DNS, hosting, payment processing), AI agent registry (every Claude/Manu/ATLAS prompt with versioning), and operational runbooks (incident response, lead handling, content workflow). Wired in a custom MCP wiki-lint tool that runs monthly to find orphan pages, dead wikilinks, and stale claims — keeps the graph clean as it grows.
◇ Decisions
Engineering choices worth flagging
Obsidian over Notion
Local files, plain markdown, no vendor lock-in. The vault is just a folder — exportable, version-controllable, and Claude can read it directly via the obsidian-vault MCP. Notion's database features would have been nice, but the lock-in cost is too high for an ops wiki.
Top-level entity foldering
WSJR/<entity>/ instead of mixing everything. Permissions naturally follow folders: someone working only on TopStocx doesn't need access to the Investments wing. Cleaner mental model, cleaner future ACLs.
Monthly auto-lint
Knowledge graphs decay. A wiki-lint script runs on the 1st of each month, checks for orphans + broken links + stale dates, and emails me the report. Same pattern as the marketing-analytics MCP — scheduled, autonomous, no manual ops.
◇ Results
What shipped
◇ Stack
Tools used
◇ Keep exploring