Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
by Patrick Lewis et al.
The 2020 paper introducing retrieval-augmented generation: a parametric seq2seq model combined with a non-parametric Wikipedia index accessed by a dense retriever, trained end-to-end. Explains why grounding generation in retrieved passages improves factual accuracy on knowledge-intensive tasks.
More resources on Retrieval-Augmented Generation (RAG)
Building and Evaluating RAG Apps
Short DeepLearning.AI course with Jerry Liu and Anupam Datta covering two retrieval upgrades over naive RAG, sentence-window and auto-merging retrieval, plus the RAG triad of context relevance, groundedness, and answer relevance, measured in TruLens.
Vector Databases from Embeddings to Apps
Explains how embeddings are formed, how similarity search algorithms index and query them at scale, and how vector stores underpin RAG. Hands-on labs build sparse, dense, hybrid, and multilingual search over real datasets using Weaviate.