Metamaterials consist of elementary, repetitive structures known as unit cells, which are used to control their mechanical behavior. Typically, geometrical parameters realize this control, and they can be tuned to satisfy specific engineering purposes, e.g. adapted stiffness or negative transverse contraction. Depending on the geometry of unit cells, metamaterials become programmable, i.e. responsive to external stimulus. In this contribution, we focus on physical stimuli like force and heat. Usually, metamaterials consist of unit cells arranged in a homogeneous layout on micro- or mesoscale. However, we intend to create heterogeneous layouts of unit cells on the same scale, where we exploit the parametric design of the cells. The effect of the geometrical parameters on the mechanical properties is directly recognizable on macroscale. This multiscale approach allows to achieve desired material behavior in metamaterials, which can be regarded as inverse design problem. The novelty of this framework is that inverse design problems can be solved with a variety of unit cell design options, even for multi-material and multi-physical cases. The design of the unit cells is obtained either automatically by topology optimization or by manual design. To capture multiscale effects of unit cells, we compute strain-stress responses for all important design and loading cases in advance. After this intensive computational phase, large data sets are created, which represent the effective behavior of the unit cell in a high dimensional space. To compress the amount of data and decrease the complexity, we use model order reduction and neural networks. By means of the resulting surrogate models of unit cells, the parameter optimization for the inverse problem becomes very efficient. In this contribution, we explain the described process chain, i.e., from the design of unit cells until the final model of the metamaterial, ready to be manufactured.