In the current version of the visual deep learning (see doc: https://doc.dataiku.com/dss/latest/machine-learning/deep-learning/index.html) of Dataiku DSS, you can only train models with one output. We support using RNN, and you can find examples in the code samples.
If you wish to train a model with multiple outputs, you can use a Python recipe. You will need to write the preprocessing and the architecture of the model. You will also need to write your implementation fo the beam search.