In this paper, we propose a novel pipeline for the generation of recommendations that utilises a transformer-based architecture for sequence-aware embedding learning and next-item prediction with a multi-objective optimisation capability in pursuit of balancing multiple, potentially conflicting recommendation goals. We demonstrate the working and utility of this pipeline in a case study in which the SASRec transformer model is trained on three distinct data sets, focusing solely on user interactions so as to demonstrate its sequence modelling capabilities. Candidate recommendation sets are generated by utilising the transformer’s item predictions and leveraging knowledge about item embeddings, after which these sets are optimised by invoking a population-based metaheuristic. Three recommendation objectives are considered, namely the accuracy and diversity of, as well as revenue generated by, collections of trade-off recommendation lists. A final recommendation list is selected by invoking a multi-criteria decision analysis technique aimed at ensuring alignment with both user-centric and business objectives. The methodology underlying this case study demonstration offers a flexible solution for developing personalised recommendation systems.