A three-dimensional extension of the Varied-Shape Voronoi Tessellation (VSVT) framework [1] is proposed for the design and reconstruction of anisotropic multiscale lattice materials. Unlike conventional two-dimensional VSVT approaches that rely on direct Voronoi generation, the proposed method introduces a convex-hull-based geometric reconstruction strategy to address the increased topological complexity and inter-RVE connectivity challenges inherent in three-dimensional settings [2, 3]. The design domain is first discretized into a structured voxel grid, where each voxel is assigned an identifier according to the optimized macro-scale design variables [4]. Voxels sharing the same identifier are grouped to form individual representative volume elements (RVEs). For each RVE, a three-dimensional convex hull wireframe is constructed from the aggregated nodal coordinates, ensuring consistent geometric connectivity across neighboring RVEs with distinct anisotropy indices [5]. This strategy avoids the need for predefined material interfaces and enables smooth transitions between heterogeneous anisotropic regions [6]. To reduce geometric redundancy while preserving convexity, a selective convex hull simplification scheme is employed. Numerical testing shows that retaining only 30% of the original convex hull facets represents the minimum threshold required to maintain a valid convex polyhedron without introducing fragmented surfaces. Subsequently, adjacent triangular facets with a normal deviation below 15° are merged, yielding simplified and topologically clean convex polyhedra. All processed polyhedra are assembled into a continuous macro-scale structure, which is finally converted into a voxel-based representation using a level-set method [7]. The proposed 3D VSVT framework enables the generation of stochastic, anisotropic, and multiscale lattice architectures with enhanced inter-RVE connectivity, geometric robustness, and manufacturability. The resulting structures exhibit disorder, non-uniformity, and aperiodicity, making the method well suited for multiscale modeling, homogenization analysis, and topology optimization of advanced architected materials [2, 4, 6].