How do you design an effective dataset curation and filtering pipeline for instruction tuning a model on specialized customer support data at Decagon?
