AI Platform Agentic System Stealth

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

Source Change
Change Detector
Impact Analyzer
Content Regenerator
Content Evaluator
Auto-approve
(mechanical fixes)
Human Review
(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.

Interested in applying this pattern to your knowledge base or content system?