The Plum Defense fine-tuning system is meant to be run on a scheduled basis with a cadence dependent on the amount of training data generated. With more training examples generated as users interact with the application, the Plum evaluator, augmenter, and fine-tuner system will re-run to continuously increase the quality of the LLM output along the metrics specific to the business case.If the LLM application has to keep up with a high amount of data drift in production, this evaluation-augmentation-retraining process could be run multiple times an hour. By eliminating the friction at each of these steps, Plum Defense allows tech leaders to make use of their LLM data flywheel.