Retrieval-Augmented Generation Architecture for Improving the Reliability of Generative Information Systems
Анотація
The paper substantiates the use of retrieval-augmented generation as an architectural approach for generative information systems operating on local, regulatory, or frequently updated data. The workflow is considered as a sequence of corpus preparation, chunk indexing, semantic retrieval, context construction, and answer generation. A practical set of reliability criteria is proposed, including retrieval relevance, evidence coverage, source traceability, and resistance to stale data. Retrieval augmentation does not automatically eliminate erroneous outputs; however, it provides a technical basis for controlling information provenance and updating knowledge without full model retraining.
Ключові слова
retrieval-augmented generation, large language models, information systems, semantic search, reliability, artificial intelligence
How to cite (DSTU 8302:2015)
Petrov M. Retrieval-Augmented Generation Architecture for Improving the Reliability of Generative Information Systems. Contemporary Science, Technology and Society : Proceedings of the International Scientific Conference / Academia Publishing Hub (London, 21 September 2026). London, 2026. P. 80–82.
