The use of Physics-Informed Recurrent Neural Networks (PIRNNs) offers a promising framework for modeling dynamic systems. They combine traditional machine learning with the system's phenomenological model by incorporating differential equations directly into the training process. This work presents the deployment of a PIRNN on a low-cost microcontroller to operate as a virtual analyzer, tested in a hardware-in-the-loop environment for PI level control of a cascaded spherical tanks system. The performance and robustness tests demonstrate reductions of over three times in computation time and the ability to provide information in the absence of measurements. These results suggest that PIRNNs can be an efficient solution for systems requiring real-time control and process analysis applications.