DutchBoy Semi is an edge-native AI diagnostic engine engineered for semiconductor front-end fabrication. Traditional preventive maintenance (PM) relies on single-variable, threshold-based alerts that generate continuous false alarms while leaving yield-impacting anomalies hidden in high-dimensional sensor data. Running directly on an Edge Server inside the fab, DutchBoy Semi maintains a continuous one-second inference cycle without network latency. The system uses Adaptive Dynamic Time Warping (ADTW) for unlabelled pattern extraction, Graph Neural Networks (GNN) for simultaneous anomaly scoring across equipment, chamber, and sensor levels, and Test-Time Training Methods (TTTM) for real-time model adaptation. By prioritizing root causes by likelihood and automatically generating structured Root-Cause and Action Plan reports, DutchBoy Semi reduces preventive-maintenance turnaround time by 90% compared to legacy threshold workflows, transforming fab reliability.