Why use this? Pre-trained rerankers work well on general text, but underperform on specialized domains where terminology, document structure, or relevance criteria differ from the training ...