In this talk, we consider the unique challenges of the biosphere that recently opened in downtown Seattle. We address two of those challenges using modern deep learning techniques: computer-vision-based plant health monitoring, and microclimate anomaly detection using autoencoders on time-series data extracted from multiple sensors. Our focus is on architecting the inference pipelines for solving these problems at scale. Specifically, we highlight the inference steps and TensorRT optimizations for AWS Greengrass ML inference.