This talk explores the interplay between physics and quantum machine learning (QML). QML has long been utilised to study quantum many-body systems. However, these systems face various challenges, such as vanishing gradients and limitations in coherence time. We will first examine how QML can be used to optimise hardware directly, creating a unique and more efficient operator capable of preparing the desired quantum state. Despite the power of machine learning, it often functions as a black-box application. This talk will also address this issue from a quantum mechanics perspective, proposing that optimisation problems, such as classification or anomaly detection, can be studied by reframing them as quantum many-body systems. This approach allows us to leverage the full range of quantum theory for data analysis, improving the interpretability of QML through quantum mechanics.