Rather than relying on pre-specified ancestry labels, which often fail to capture the full genetic heterogeneity of diverse populations, SPLENDID treats ancestry as a continuum and improves polygenic risk score (PRS) prediction for biobank-scale datasets.