This document discusses using locality-sensitive hashing and minhashing to efficiently perform similarity searches on big medical record datasets. It proposes pre-computing clusters of similar patients and storing them in separate files, along with centroid files. To find similar patients, it loads a patient, reduces their dimensionality, compares them to centroids, loads the closest cluster, and compares them to patients within that cluster. This approach allows performing similarity searches in about 100 milliseconds using only 20GB of storage and 20MB of RAM on a typical laptop, providing an efficient solution for big data similarity searches.