Retrieval-augmented generation
RAG combines information retrieval with generated answers. Access to sources does not remove the need for quality and permission design.
Retrieval-augmented generation (RAG) combines fetching relevant information with generating an answer through a language model. Retrieved content is supplied as additional context. The model need not be retrained for every new document.
A knowledge assistant example
In a hypothetical business, the assistant first finds an approved work instruction and then formulates an answer linked to its source. Search might use keywords, vector similarity or a combination.
RAG does not guarantee a correct answer. Incorrect, outdated or unauthorised passages can misdirect the output. Even a suitable source can be summarised incorrectly. Source selection, access rights and faithfulness to the evidence therefore need separate checks.
Sources
Related terms
- EmbeddingAn embedding represents content as a numeric vector. The model and comparison method determine which similarities become visible.
- Vector databaseA vector database stores and searches vector representations. A similar match is not automatically a valid or permitted answer source.
- GroundingGrounding links an AI answer to verifiable evidence. A source link alone does not show that every claim is supported.
- Model cardA model card documents an AI model’s purpose, foundations, evaluation conditions and known limitations for a particular version.
- AI readinessAI readiness concerns the foundations of an AI project: objectives, usable data, skills, responsibilities and an operating workflow.
- Context engineeringContext engineering prepares information for individual model calls: relevant instructions, messages, sources and current states.
Articles on this topic
AI & Automation with ASES
AI assistants and automated workflows