Luminary Life
CAIO Platform
A multi-tenant fractional Chief AI Officer platform where Postgres row-level security is the security boundary, not middleware.
01 The story
Live at caio.luminarylife.ai. This is the platform for running AI strategy across a portfolio of client companies, and it is built security-first: the database itself enforces who sees what, the service role is confined to one audited module, and the audit trail is append-only by construction. Scores are never computed at render time; they are versioned records tied to a methodology version and evidence with explicit confidence tiers, so a score from March means the same thing in September.
02 What went into it
- Row-level security as the boundary on every tenant-scoped table
- 18-driver scoring engine with immutable methodology versioning
- Evidence tiers: system-pulled, operator-verified, founder-attested with 90-day expiry
- Append-only audit via a SECURITY DEFINER RPC
- Operator, client, partner, and admin route groups with scoped access
- Claude-scored driver evaluation with persisted explanations
03 Claude in production
Every AI call runs server-side under the Lumen governance layer: no client keys, cost telemetry on every call, and a human confirming every consequential write.
04 Stack and capabilities