Physics-informed neural networks (PINNs) are deep learning neural networks explicitly conceived as an alternative solver of partial differential equations with respect to standard numerical techniques. PINNs offer some unique features, such as the capability of constraining the solution with internal or external and local or integral information, allowing to take into account uncertainty of this information. They can also be constrained with incomplete physics equations, allowing the development of modelling tools. Therefore, they offer the possibility of developing a unique framework, which permits to combine physics and data. In this work, their potential has been investigated by applying them to one of the most important inverse problems in tokamaks, the plasma equilibrium reconstruction. More specifically, an advanced PINN-based equilibrium reconstruction method has been developed that combines multi-diagnostic constraints with high-fidelity physics modelling of the measurements, able to take into account both non-linearities and relativistic effects. All the relevant diagnostics have been included in the study, confirming the potential of the technology to perform also integrated data analysis. A series of numerical tests, performed with the help of the Tokalab platform, have proven the quality of the results in cases, for which the right solution is known. After this validation, the developed tools have been applied to analyse various Joint European Torus (JET) discharges, with particular attention to high performance experiments in DT. A detailed comparison with the reference inversion codes used on JET (EFIT, EFTP and EFTF) is reported together with diagnostic ablation tests, confirming both the accuracy and the reliability of the approach. The obtained performances motivate various future developments such as the implementation of multi-fluid magnetohydrodynamic equations, plasma dynamics reconstruction, and acceleration schemes to reduce the computational times.