I have answer to the Android part now. I have mobile, especially Android development as background, so maybe be I am biased a little bit. But generally, I consider edge ML(including mobile), very interesting from various perspectives. As user/consumer enjoy AI/ML applications without compromising his/her privacy by doing inference, even additional transfer learning on local device, without cloud. Also low latency makes it a nice user experience (I was getting 3 FPS on live camera stream with an older Android without GPU acceleration with Tensorflow Lite on PoseNet). Mobile SoC having dedicated "neural engines" (like Apple Ax or M1 processors, or considerable effort by Si-Five with Chris Lattner) are for me a strong signal for an demand. Other is also an engineering perspective, where you can have an advantage using light Kotlin based stack instead of general purpose heavy weight frameworks. Also speaking of edge, imagine embedded devices with smaller chips. JB having Kotlin Native, powerful compiler with plugins...