The genotype similarity search (GenoSiS) workflow. (A) Starting with a VCF and recombination map, GenoSiS generates a haplotype positional encoding vector for each sample. (B) We then train an embedding model to minimize the difference between encoding and embedding distances. (C) Using the trained model, we create embeddings for all sample haplotypes and index them with Intel's SVS. (D) The index is queried using a sample's embeddings to find its nearest neighbors, and those results are aggregated to determine the sample's cohort.