Retrieval Augmented Generation (RAG) is a framework of enhancing the output of generative models by providing external knowledge during generation. In the setting of question answering, RAG provides a large language model with a set of documents related to the specific query, allowing the model to generate answer that is more accurate and comprehensive.
In this project, we build a RAG pipeline that takes in a query from user, retrieves relevant information from the database, and output a summarization by a large language model. More specifically, we focus on building a model that can retrieve the information fast (sub-seconds) and can also filter out irrelevant information after retrieval. We also provide a framework using RAGAs to evaluate our model.