Many advanced sensors are capable of producing extremely large and continuous data streams. Hyperspectral imagers, microphones, high-resolution cameras, and 3D scanning devices can easily generate gigabytes of data per day, making it impractical for many wireless sensor platforms to stream all collected data to the cloud for analysis. Furthermore, in some sensor deployments, privacy concerns may restrict the resolution or content of data leaving the devices and being routinely stored in the cloud. To address this situation, sensor platforms must reduce or transform the data in situ, sending only the analyzed results to a central server. In designing an open sensor platform capable of leveraging advances in machine learning to support edge computing, several challenges arise, including resilience, performance isolation, and data privacy. This paper describes the architecture of the Waggle platform developed at Argonne National Laboratory. As an open platform, Waggle supports a wide range of sensors, including experimental sensors to measure airborne pollutants such as hydrogen sulfide and ozone, as well as cameras intended to detect urban flooding and automobile traffic. The Waggle platform is used by the Array of Things, a National Science Foundation project to deploy 500 sensor platforms in the city of Chicago, beginning in mid-2016.