Two studies highlight how edge computing architectures are moving from benchmarks toward resource-constrained deployments with direct healthcare relevance. One report describes an “8-kilobyte” physics-informed edge AI model that compresses precision irrigation inference onto microcontrollers while preventing crop dehydration via embedded biophysical guardrails. A separate report validates a federated-learning, homomorphic-encryption-and-blockchain approach that achieves high cardiac classification accuracy on an embedded device while cutting inference latency by 52%. While these studies span agriculture and cardiology classification, they both emphasize on-device constraints—memory footprint, latency, and privacy controls—as gating factors for real deployments. For biotech and digital-health teams, the combined signal is that model compression plus on-device trust infrastructure is becoming central to practical AI integration in clinical workflows.