The lost foam casting process is a form of evaporative-pattern casting selected to manufacture automotive parts, where a polymeric foam pattern, covered by a refractory material permeable to gas and surrounded by sand is replaced by a flowing hot liquid metal. The basic model consists in pouring the liquid aluminum into the mold cavity, that decomposes the foam into liquid polymer and gaseous byproducts, and forming a decomposition layer between the advancing metal and the receding foam pattern casting. Modeling such a multi-scale problem can lead to undesirable computational cost due to the high coupling between the physics involved and the wide ranges of scales to consider. Consequently, we propose an advanced flow modelling technique, with the least computational cost, developed for an accurate estimate of the evaporative front kinetics in random 3D topology. This new technique consists mainly in treating the advection of the liquid metal front, captured using a levelset method, along a unitary vector field, similarly to a neutral line direction in structural dynamics. The advection velocity is explicitly expressed in terms of the process parameters and geometric variation and was derived by coupling multiple phenomena: (i) 1D heat transfer through the vapor layer, (ii) foam phase change into residues (iii) 2D mass transfer of the gaseous residues through porous media. From (i) and (iii) emerges a governing geometric parameter: the local surface/perimeter ratio, which alters the mold filling speed at each time step. A front flagging method was adopted to treat multiple front cases based on diffusion operators. This global model proves to be a good candidate to predict complex flows. However, it was necessary to enrich the explicit expression of the velocity by considering local distortion of the vaporizing front. Indeed, we show, using a more refined local model consisting of the three phases (metal liquid, decomposition layer and EPS), that some additional heat transfer occurs between the vapor layer and the polystyrene beads, located in a thin undercut layer close to the walls, which in return increases drastically the surface of mass exchange of vapor through the walls. Both models were solved numerically, using our finite element library. The final solution shows a good prediction of the experimental front kinetics as well as complex phenomena such as fronts merging in 3D highly curved industrial parts.