The paper presents a brief overview of the key features and characteristics of neural networks in relation to microcontrollers. We examine two popular and affordable hardware platforms, ESP32 and Arduino Nano, which are commonly used for the implementation and testing of neural networks. Thus, an analysis is conducted on both a low-end and a high-end microcontroller to assess their efficiency and performance in a variety of contexts. The implementation methodology of neural networks on microcontrollers is the process of integrating the weights and biases that have been trained into real applications on the target hardware devices. Furthermore, the paper addresses quantization methods, which are essential for reducing the size and complexity of the models, thus facilitating their implementation on microcontrollers with limited resources. A reliability study on the system was also conducted, during which the requisite calculations were performed on the components. This resulted in a final reliability value of 0.84. Additionally, a system analysis was performed, entailing the realization of a Markov network and transition matrix. This allowed for the observation of the transition behavior from one state to another within the entire system.