Retrieval Augmented Generation (RAG)

Pattern that retrieves external knowledge and provides it as context to a model.

1.
Notion AI uses RAG to ground answers in a company's own wiki pages - retrieving the top-k relevant documents via semantic search and including them in the prompt before generating a response.
2.
Morgan Stanley built an internal RAG system over 100,000+ financial research documents so advisors can ask 'What does MS think about NVDA's data-centre outlook?' and receive grounded, cited answers.
3.
ServiceNow's AI platform uses RAG to ground its helpdesk agent on live knowledge-base articles, reducing hallucination by 40% and cutting average ticket resolution time from 12 minutes to 3 minutes.

Retrieval Augmented GenerationRAG