The pipeline begins by selecting diverse top RBX1 seed binders from the combined input dataset using Pareto-based NSGA-II multi-objective ranking, ensuring that multiple high-quality candidates are retained instead of relying on a single heuristic score. Seed selection is based on structural and sequence-level metrics such as ipSAE, iPTM, complex pLDDT, interface contacts, epitope coverage, PRODIGY ΔG, SAP score, and surface hydrophobicity, with strict RBX1-specific thresholds applied to eliminate weak binders with poor interface quality or epitope engagement. Each selected seed is then optimized independently on its fixed backbone structure, and ProteinMPNN performs constrained redesign to generate sequence variants by mutating interface residues and liability-prone positions. During each optimization epoch, four redesign strategies are explored: interface refinement, surface cleanup, mixed exploration, and mixed exploration with solubility bias. Generated variants are first evaluated using sequence-level metrics including ProteinMPNN scores, global scores, PLM_Sol solubility predictions, and developability descriptors. Promising variants are then subjected to structure-level rescoring using tools such as Boltz, ipSAE, PRODIGY, interface contact analysis, buried surface area, SAP score, and binder surface hydrophobicity. A second NSGA-II selection step identifies elite variants across epochs while preserving diversity across designs. The pipeline ultimately produces optimized candidate binders for each seed family, along with a combined final shortlist of high-confidence RBX1 binders selected across all optimized seeds.