This work presents a framework that leverages Machine Learning (ML) to design Functionally Graded Meta Materials (FGMMs) with objective mechanical performance. An ML-based inverse generator is developed to facilitate the creation of metamaterial cells based on target property inputs [1]. A design framework with a rough-mesh Topology Optimisation (TO) was first created to determine the initial material distribution across the design domain. This is followed by clustering the mesh elements based on the density and stress status from the TO result using k-means clustering. As shown in Figure 1, for each cluster, the optimal mechanical properties are identified and then mapped directly to the metamaterial cells via an ML-based inverse generator. The effectiveness of this FGMM design framework is demonstrated through real-world engineering applications, specifically targeting mechanical deformation minimisation and functional requirements. By comparing this approach with a traditional de-homogenisation method, the proposed framework demonstrates its advantages in reducing computational costs while achieving superior design performance.