Initial version of representation learning model for eCLEF18-Task1. The model is able to distinguish between correct line-ICD10-code pairs (originating from the training data) and randomly sampled corrupted pairs.