Under the guidance of Professor Minjia Zhang at UIUC, I collaborated with Xiao Wang and others to develop MedCite, a framework designed to enhance verifiability in medical question-answering systems. MedCite improves citation generation by leveraging dense and sparse retrievers to recall and rerank relevant articles, which are then used for retrieval-augmented generation by large language models (LLMs).

Additionally, MedCite includes a post-generation retrieval method that generates accurate citations for the produced text. An automated evaluation module further assesses citation quality using precision, recall, and uncertainty metrics. Our work has demonstrated significant improvements in citation quality, confirmed by feedback from medical professionals.