My research focuses on safe and resilient AI for autonomous systems. I design detector-agnostic, sequence-aware run-time monitors that introspect internal neural representations of LiDAR-centric perception to predict, localise, and explain false negatives/positives. My work couples uncertainty from activation patterns with temporal consistency and multi-modal cues, translating introspection signals into policy/SOTIF-compliant safety actions (alerts, MRM). Beyond driving, I explore lightweight anomaly detection and HAR on wearables, extending these principles to broader safety-critical sensing.