Lattice structures - inherently multi-scale by design - have attracted interest from diverse fields owing to its promise for high strength-to-weight ratios as well as improved energy absorption or wave attenuation performance. Several numerical methods to design multi-scale lattice structures, based on topology optimisation (TO), have been developed, e.g., inverse homogenisation and dehomogenisation. Our previous study [1] leveraged machine learning (ML) -based inverse generator along with TO, which efficiently optimised 2D graded lattice structures exhibiting comparable or superior mechanical performance to the state-of-the-art. In this work, we will compare and contrast the extensions of the proposed method to other established approaches considering several aspects including computational efficiency, design flexibility and manufacturability. The novelty of this work is in the incorporation of orthotropic material properties and geometric non-linearities, advancing on [1], to ensure manufacturable fibre reinforced additive manufacturing (FRAM) solutions. An extension into three dimensions is proposed to include plate-based lattices, which have been demonstrated to achieve high strength-to-weight ratios among lattice families, albeit typically at a mono-scale. Preliminary results confirm the performance benefits of these in the multi-scale paradigm. Besides the numerical investigation, structures are fabricated in short carbon fibre reinforced polyamide, utilising Anisoprint Composer A3. Experimental validation is conducted in three point bending test based on ASTM D7264/D7264M-15, where the strain field is measured by digital image correlation (DIC). This study aims to pave the way for enhancing the versatility of the lattices by overcoming typical challenges in the field, such as geometrical nonlinearity and manufacturability.