This project implements a supervisor-worker multi-agent pattern for real-time fraud detection. The system runs entirely locally (no cloud ML services), using Claude's API only for agent reasoning ...