ContentGuardian
An auto-healing content platform that detects when source knowledge changes — and fixes itself.
The Problem
Certification programs and knowledge bases go stale the moment underlying sources change. Keeping them accurate requires manual review at scale — expensive, slow, and error-prone.
The Solution
ContentGuardian runs four composable AI agents in sequence: detect changes in source documents → analyze which content is now stale → regenerate accurate replacements → evaluate quality with an LLM-as-judge → route through human approval or auto-approve. Every step is observable, every change carries full provenance.
Architecture
(mechanical fixes)
(substantive changes)
Key Capabilities
4-agent agentic healing pipeline
Human-in-the-loop governance with policy-based routing
LLM-as-judge evaluation (groundedness, accuracy, pedagogy, hallucination risk)
Full provenance — every version traces to its source
Dual-database (SQLite dev / Supabase + pgvector production)
Zero-config demo mode with deterministic mock provider
Tech Stack
Next.js 16 · TypeScript · Mastra · Google Gemini · Drizzle ORM · SQLite (dev) · Supabase + pgvector (prod) · Tailwind · Vitest
Outcome / Status
Functional MVP deployed at [demo URL — TODO]. Demonstrates production-ready agentic architecture with full observability, governance, and reproducible demo. Actively seeking pilot organizations in certification, training, or compliance content.