Skip to content

El Brain-o, second brain for Kabbalah Centre ops

A system that reads everything, thinks about what matters, and drafts the work.

Twenty-nine small workers watch the meetings, emails, and class calendar, turn the signal into tracked tasks and ready-to-send drafts, and hand every outbound message to a human for one tick of approval. Here is the whole machine, made visible.

Live system map interactive
SIGNAL INTHE BRAINDRAFTS OUTANALYZERelevance filterdomain to title to AIConfidence tiersHaiku to Sonnet to OpusCost meter$4/day soft capCachesame input, one callFireflies.aiMeeting transcriptsEmailOutlook / GmailClass Ops v2Notion, read-onlyClaudeAnthropic APIGeminiSecondary classifierOutboxhuman approvesThe brainSystem B writesMorning Brief5am ET digestMission Controllive dashboardhuman ok

Tap any node to see what it does. Amber flows in, the brain analyzes, green drafts flow out, and nothing leaves without a human tick.

29
scheduled workers
2
Notion workspaces
$4
daily AI budget
1
human approval gate
01

Signal in

Meetings, emails, and the live class system pour in raw. A three-tier filter keeps only what is actually about a Kabbalah Centre class, so the brain never drowns in noise.

  • Fireflies.ai
  • Email
  • Class Ops v2
  • Claude
  • Gemini
02

The brain analyzes

Cheap models handle the routine; only genuinely hard calls reach the expensive one. Every call is metered against a hard daily budget, and the same question inside a day is answered once.

  • Relevance filter
  • Confidence tiers
  • Cost meter
  • Cache
03

Drafts out

The system writes reminders, recaps, and posts, and files what it learned. But it never sends. Every outbound message waits in one queue for a human to tick approve.

  • Outbox
  • The brain
  • Morning Brief
  • Mission Control

The one rule that makes it safe

No worker ever sends anything on its own. Every email, post, and reminder is a draft until a human ticks approve.

The control flow lives in plain code, not in a model deciding what to do. AI is used only at three narrow points: writing copy, sorting the genuinely ambiguous, and pulling structure out of free text. Everything else is deterministic, logged, and reversible.

Go deeper

Three ways into the system