In this situation I'd go for two approaches: - have your labeling process write to a dataset kedro can read - have the output of your labeling process passed in to kedro run --params