Would love to see this connect to a smartphone running a matching app, doing the inferencing on the phone so you wouldn't have to bill for increments. It will be a few more years until that can be done in a low-enough latency way (improved models, more compute and memory available).