Massive amounts of data being created at the edge, uploading this data to the cloud for real-time analysis is neither financially nor technically feasible. Extracting meaning from this data is computationally intensive and expensive, limiting adoption of edge AI to a few high-value applications. After years of research by leaders in computer vision and machine learning we have developed technology that is two to three orders of magnitude more efficient than other high-accuracy deep learning systems. Our patent pending technology evolves the state of the art with a combination of optimization, sprinkled with unique insight and just a little magic. Video is our first application: the system is trained in the cloud by ingesting video where objects and activities of interest have been annotated. Learning is distilled into ultra-efficient software that runs on fixed, body-mounted and vehicle-mounted cameras. We can enable analytics on most professional IP video cameras with just a firmware update. Older cameras can be retrofitted with a low-power embedded processor, such as those from our partners NXP and Ambarella. Our software-only system gets smarter over time: edge cases (where the system is not very confident that it knows what is happening) are captured, uploaded to the cloud for analysis, and the learning distilled into ever smarter software. We work with more than just video: infrared, thermal, radar and lidar play important complimentary roles in helping to establish accurate, detailed situational awareness across a wide variety of environmental conditions. Each sensor detects, classifies and tracks multiple objects locally in real-time. This metadata is then push to the cloud, or a central ECU in the case of a vehicle, where it is fused to establish ground truth, detect behavior, track objects across multiple sensors, and identify anomalies.