Countless microcontroller-based IoT condition monitoring sensor solutions in machines and plants only capture raw data of motors, bearings, fans, pumps, etc. and transfer these data to external cloud services to determine the current state of operation. Such kind of sensor application, that use AI algorithms in the cloud generate an unnecessary huge amount of communication data overhead in internet communication and in some cases unacceptable latency problems. AI algorithms concepts, such as Supervised Machine Learning, consist of two individual functions: a learning phase for build a machine learning model and an inference phase to use this model for regression or classification. The speaker in this session shows an example of how a supervised machine learning model is created in a public cloud and then used on a microcontroller directly in the sensor to carry out state determinations in a real-time inference phase.