Driven by the maturation of additive manufacturing technologies, topology optimization has experienced exponential growth since its seminal work in 1988. Numerous approaches have been proposed for solid mechanics, fluid mechanics, and, to a lesser extent, multiphysics applications. However, most existing work focuses on minimizing properties defined in a single physical domain (e.g., minimizing structure compliance or fluid energy dissipation). Considering nuclear industry applications, multiple physical properties must be considered simultaneously in design, motivating the use of a multi-objective function. Treating such multiphysics cases often requires intrusive modifications to existing solvers or reliance on open-source platforms (e.g., FreeFEM, FEniCS). The proposed work presents a non-intrusive and user-friendly framework for industrial applications, in which the optimization solver is separated from the PDE-physical solvers. Users can independently integrate different software to compute the physical fields needed in the optimization process. In the present implementation, Computational Fluid Dynamics (CFD) is solved using Simcenter STAR-CCM+ and Finite Element Analysis (FEA) is performed with Cast3M. Their coupling and the solution of numerical schemes are managed by a Python code. The multi-objective function to be minimized is defined as the weighted sum of quantities in both subdomains. Inspired by a promising approach result of a previous PhD project, a multiscale level-set topology optimization method is developed to improve local performance. Unlike methods that rely on a single mesh for the fluid, solid, and level-set field, the proposed approach employs three distinct meshes: two regenerated at each iteration for CFD and FEA, and a fixed structured mesh for the level-set scalar field. This allows each physics to be solved on a mesh adapted to its needs, while enabling efficient finite-difference schemes for solving the Hamilton-Jacobi equation. Architected materials are incorporated into the optimized designs. The method is validated using established 2D and 3D cases from the literature.