This repository contains resources and code for enhancing Retrieval-Augmented Generation (RAG) systems using a novel approach termed LOTR (Lord of the Retriever). The primary focus is on addressing the 'Lost in the Middle' (LIM) challenge in RAG systems, particularly in the context of medical/healthcare data.
Advanced Retrieval Techniques: Utilizes multiple vector stores and the Merge Retriever approach to efficiently retrieve relevant documents. LOTR - Merger Retriever: Combines results from various retrievers to form a comprehensive, relevant document list. LongContextReorder (LOTR): Reorders information to ensure equal attention to all parts of the text. Domain-Specific Embeddings: Incorporates specialized embeddings for medical and healthcare-related data.
Colab walkthrough for LOTR
For a deeper dive into the cutting-edge technologies of FLARE, and to access detailed technical knowledge, check out our Medium Blog.