In this talk, we present an efficient approach for solving multiscale PDEs with high-contrast coefficients, combining multiscale ideas with domain decomposition techniques on irregular subdomains. The method is based on the Virtual Element Method and uses generalized multiscale constructions to build reduced spaces from local spectral problems, capturing fine-scale features efficiently. These spaces are then incorporated within a two-level domain decomposition framework to ensure robustness and scalability. We also briefly discuss time-dependent problems. Numerical results illustrate the accuracy and efficiency of the approach in heterogeneous multiscale settings.