The talk will showcase some of the Machine Learning (ML) methods being applied in B physics spectroscopy. Using the data collected with the CMS experiment and the Bs→psi(2S) phi decay as a playground several most commonly used algorithms will be examined and compared while searching for a well-known intermediate psi(2S) resonance. The main emphasis will be given to a proper understanding of the given ‘signal vs. background’ classification problem with a thorough discussion on possible pitfalls which could arise from both physics and ML sides. The motivation behind the talk is not just to present how ML can be applied in B physics and how it could significantly improve the analysis performance. It is of more importance to emphasize the bridge one could build between physics and ML to complement each other. Namely, to show how one could deploy physics knowledge to a given ML model to modify it to one’s needs and thus obtaining more physically motivated results and vice versa - how a properly trained model is in fact highly interpretable in a physics sense, which allows one to gain better insights into the problem.