Accurate prediction of Critical Heat Flux (CHF) is essential for the safe and efficient design of thermal systems, particularly in nuclear engineering. Over the years, numerous prediction models have been developed, including empirical data-driven correlations and mechanistic models. More recently, machine learning (ML) and artificial intelligence (AI) techniques have emerged as potential alternatives to traditional prediction methods. This thesis proposes a Physics-Informed Neural Network (PINN) that incorporates CHF physical knowledge into the learning process in order to improve predictive accuracy and generalization capability. To overcome the absence of a complete system of governing equations describing CHF and the well known limited accuracy and validity range of existing physical models, a K-Nearest Neighbors (KNN) based prior is developed.