Identifying low-GWP refrigerant mixtures for specific heat pump applications requires coupling working fluid selection with cycle performance evaluation. This work proposes a framework which connects refrigerant screening at the molecular level with heat pump system simulation for a dishwasher application. A multi-objective genetic algorithm simultaneously optimizes binary and ternary mixtures from 27 pure substances, using coefficient of performance (COP) and global warming potential (GWP) as competing objective functions. Each candidate is evaluated in a vapor-compression cycle model that includes pressure drop estimation at an evaporation temperature of 276.15 K and condensation temperatures ranging from 303.15 K to 348.15 K. Mixture properties are calculated from the multi-fluid mixture model; where adjusted binary interaction parameters are unavailable, the model is supplemented by COSMO-SAC-dsp as gE-model in the theoretically based departure function for predictive capability. The algorithm converges to R-290/RE-170 (0.854/0.146 mol/mol) as the optimal blend, achieving a COP equivalent to the pure R-290 reference. The near-pure behavior of the optimum reflects the application-specific constraints that penalize temperature glide, limiting the design space for mixture benefits. The methodology is transferable to applications where glide-tolerant heat exchangers may unlock larger mixture advantages.