optimal_dragonfruit
The scope of this project was to perform an end-to-end, adversarial workflow to identify and correct failures in a large language model. The project's goal was to "stump" the model, verify the nature of its failure, and then generate a "golden" data set for fine-tuning.