You'll have to manually compute the mean after retrieving the nearest neighbors. Automatic fast averaging could be an interesting use case for clustering or using a vector database to generate training though.