Please see my notes in this thread -> https://195.201.255.233/forums/topic/mdma/
This endeavor fails feasibility until the SDK is capable of returning a confidence number of how well a scan matches a built model. You can have the greatest model in the world built around MDMA, but if a user scans a banana the SDK is basically unable to tell the application that the provided sample does not match any of the loaded models. It’s pointless to proceed until this is in place, unfortunately.
