This seems pretty cool -- but one issue to me is that (similar to the chasm that exists in low-code app building once the magic doesn't suit you) that if I alre

This seems pretty cool -- but one issue to me is that (similar to the chasm that exists in low-code app building once the magic doesn't suit you) that if I already have the skills to create a mobile app that integrates tensorflow, I probably also have the skills to train my models. It would be cool if feature-extraction (image pre-processing and first network layer(s)) could run on the front-end, and the rest of the network/search on the back-end, similar to how distributed speech recognition works. Then I could use a canned lib on the device that integrates w/the camera, and get my results via a websocket. (Of course, I could still run everything on the client still as well.)