Lipoprotein(a) [Lp(a)] is a highly atherogenic, pro-thrombotic lipoprotein and a major risk factor for cardiovascular disease for which highly potent therapeutics are under evaluation. Lp(a) consists of a low-density lipoprotein (LDL)-like particle covalently bound to apolipoprotein(a) [apo(a)]. Apo(a)’s size variation, a major driver of Lp(a)’s high heritability (~80%), is largely determined by a variable number tandem repeat (VNTR) comprised of two-exon repeat units that together account for approximately 70% of the coding sequence. The extreme VNTR length and repetitive structure is difficult to resolve and phase using both short- and long-read sequencing technologies. We are developing a computational framework to estimate haplotype-informed KIV-2 VNTR length across diverse populations, with an emphasis on scalability for large epidemiological datasets. In this framework, we integrate multiple forms of genetic ancestry (e.g., local ancestry and identity-by-descent) to improve phasing accuracy in this complex genomic region where in-phase variants can modify the effect of VNTR length on Lp(a) levels. We are applying this framework in a consortium of ancestrally diverse studies to capture population-specific variation in VNTR length and to generate more accurate, ancestry-aware polygenic risk scores (PRS) for Lp(a). Our open-source Python package, Genomic Repeat inference from Depth (GRiD: github.com/caterer-z-t/GRiD), provides an accessible, flexible, high-performance solution for modeling the genetic architecture of Lp(a). In preliminary analyses, GRiD VNTR-length estimation correlated with Lp(a) levels (R²=0.165;p<0.001), and we expect PRS performance to be highly accurate across population subgroups when combined with LPA variants (results forthcoming). By uniting VNTR-length inference with population-structure information, GRiD facilitates improved genetic risk predictions in large-scale studies and supports downstream translational analyses of cardiovascular risk.