Suppose there are few very huge shuffled datasets of embeddings. DBSCAN (doesn`t matter, what algorythm, nut number of clusters is unknown) is used to clusterise them by cos
