A major issue in driver identification is lack of enough data from the target group of drivers. In this work I use Triplet loss to train neural nets that learn to discriminate between small chunks of driving data. Through this we significantly reduce the amount of data needed to perform accurate driver identification or verification.