Protein assemblies facilitate many essential roles in biological systems, from structural support to communication and catalysis. Designed modular protein assemblies hold significant promise for applications in therapeutics, nanomaterials, and synthetic biology. One challenge is the design of asymmetric interfaces to control the spatial and geometric properties of these assemblies. Many current examples rely on overall symmetries to form large assembly structures. The design of specific protein heterodimers capable of modular extension and functionalisation aims to address these issues. The designed helical repeat proteins (DHRs) represent an ideal modular protein system. Inspired by their intrinsic stability, we produced split-DHR (sDHR) protein heterodimers. Three novel split-DHR (sDHR) heterodimers—sDHR49, sDHR64, and sDHR79—are expressed and characterised via bacterial expression in E. coli, and a further two sDHRs are found to bind in metal affinity pull-down experiments. The split-DHR proteins are stable, soluble as monomers, and have binding strengths in the hundred-nanomolar range in SPR binding assays. Good orthogonality between the heterodimeric interfaces of sDHR49 and sDHR79 is observed, although some cross-reactivity is observed between sDHR64B and sDHR79A. This thesis also aims to improve experimental success rates of de novo heterodimer design and assess the possibilities of incorporating novel machine learning tools into pipelines for heterodimer design. Using experimental data from the split DHR dataset and a larger heterodimer dataset from Sahtoe et al., we performed a retrospective analysis of the Rosetta protein design method and AlphaFold2 (AF2). We can retrospectively achieve binding design success rate improvements of up to 5-fold when thresholding by Rosetta and AF2 score term values, derived from an ROC-curve classifier informed by the experimental data. The generalizability of these recommendations requires further assessment with larger experimental datasets but provides promising preliminary recommendations. This work also displays the functionalization capabilities of the sDHR proteins through design of an sDHR-ligand display complex with two biological proteins of interest for red blood cell (RBC) production, SCF and EPO, where EPO functionality is incorporated through a previously designed EPO activity mimic AR3. A non-canonical amino acid, L-HPG, is then incorporated into the sDHR79B protein, showcasing its potential for future use in click-chemistry reactions for immobilization to surfaces relevant to the production of RBCs. Finally, extended sDHR repeat proteins were designed using the Elfin software for modular protein design and successfully expressed and purified from E. coli. Negative stain EM imaging provides promising initial evidence of assembly formation for the sDHR interfaces in these extended modular repeat proteins. By offering new heterodimeric building blocks and demonstrating their modularity in extended structures, this work advances the development of asymmetric protein assemblies. We also show that integrating Rosetta design methods with AF2 enhances predictive accuracy for heterodimer designs, though further refinement of interface-specific thresholds is essential.