A domestic research team has developed next-generation databases technology to improve the answer accuracy of enterprise AI agents. The method reduces hallucinations in Generative AI by handling semantic information in documents, relationships among entities, and table-form data together in a single system.
A research team led by Kim Min-su, a professor in the School of Computing at KAIST, said on the 19th that it developed the integrated databases management system AkasicDB and the retrieval-augmented generation (RAG) technique Omni RAG based on it, in collaboration with the faculty-founded startup Graphy. The research results were presented as a demo paper at ACM SIGMOD 2026, an international conference in the field of databases, held on the 2nd.
RAG is a technology that has AI search external documents or data and then generate answers based on them. Its use is increasing for enterprise AI agents, but actual corporate data are dispersed in various forms—documents, tables, relationship information—creating limits to AI's comprehensive understanding. In this process, hallucinations can occur in which answers are generated with insufficient grounds.
Conventional RAG generally converts questions and documents into vectors to find semantically similar documents. However, there were constraints for complex queries that also need to filter by specific periods or types and simultaneously grasp relationships among entities such as people, corporations, and products.
AkasicDB is designed to execute vector DB, graph DB, and relational DB functions in an integrated manner within a single databases management system. Users can express queries that combine vector search, graph traversal, and relational filtering as a single SQL/GQL query, and AkasicDB processes it with a single execution plan.
Built on this, Omni RAG simultaneously leverages semantic information in documents, relational information in knowledge graphs, and structural conditions in table-form data. The team said this approach can reduce the likelihood of hallucinations by prompting AI to answer based on more specific evidence.
In experiments, complex search queries that took up to 21.3 seconds on existing systems were processed in under 1 second, showing more than a 20-fold performance improvement. Omni RAG increased answer accuracy by up to 78% compared with conventional RAG.
Professor Kim Min-su said, "For AI agents to use corporate data accurately, infrastructure is needed that can process vector, graph, and relational data together in a single system," adding, "AkasicDB can be used in fields where reliability is crucial, such as defense, manufacturing, finance, law, and science and technology."
References
ACM SIGMOD 2026, DOI: https://doi.org/10.1145/3788853.3801609