Do you mean other than real-time inferencing, which can be implemented using Edge Devices? This way, we should have a mother model to be trained once and use edge for real-time inferencing. Meanwhile, we can use continuous learning methods for retraining the main model.