Most industrial systems involve multiple interacting physical phenomena (mechanical, thermal, chemical, electromagnetic, etc.), often strongly coupled. While there are many solvers for each physics considered independently, a completely new code is often required when dealing with specific multiphysics problems. Multiphysics solvers can be classified into two main approaches. Monolithic approaches solve all physics simultaneously, while staggered solvers process each physics separately, transferring information between them at each time step. The latter facilitates the reuse of existing codes. However, limited work has addressed code coupling aspects; the preCICE library is a notable exception, though it is currently restricted to problems involving no more than two physics. Here, we use an approach based on the LATIN-PGD solver whose modularity is handled by its concept of interface between physics. This iterative algorithm relies on the alternative resolution of a local coupled system of equations (during the coupled stage) and of global independent decoupled problems (during the decoupled stage). It enables the coupling of models of different natures by solving them separately at the decoupled stage and projecting at the coupled stage all the physics onto the same domain, that can be interpreted as a space of accommodation. This one allows physics defined with different models to interact with each other. The decoupled nature of the algorithm allows to deal with different reduced-order model for each physics, providing flexibility, especially when dealing with different temporal and spatial discretisations. As a demonstrative example, we focus on non-linear thermo-poroelasticity problems. The novelty herein is to investigate the coupling of physics with different natures of of representation, for example considering different temporal and spatial discretisations. Furthermore, we will present results on the coupling of data-based and purely physical models.