GOAT OPS — Daily AI Ops Briefs
Every morning Claude reads yesterday's numbers across 12 dealerships and posts what needs attention to Slack.
An ops report system built like production, not a prototype. Deterministic synthetic data (14-day × 12-store, seeded, with weekday patterns and correlated funnels) made iteration safe before touching real numbers. Ingestion into Supabase is idempotent, Claude output is JSON-schema-checked with a bounded repair loop (max two attempts) before falling back, and every run is logged with tokens, cost, and checksum. Make.com orchestrates a master dispatcher plus reusable child runners.
Technologies
Key Details
- Anti-gaming LLM validation: action-item allowlist + JSON-schema checks
- Bounded repair loop — max 2 LLM attempts before deterministic fallback
- Full run ledger in Supabase: tokens, cost, checksum per run