AI & Enterprise
KAIST develops integrated DB to cut AI hallucinations, boosts accuracy by 78 percent
Researchers at the Korea Advanced Institute of Science and Technology (KAIST) developed a database technology that analyzes documents, data and relationships among entities at once to reduce hallucinations in enterprise AI agents. KAIST said the system, called AkashicDB and its OmniRAG method, integrates vector, graph and relational databases into one DBMS. It said answer accuracy improved by up to 78 percent and complex search performance by more than 20 times, cutting response delays and LLM token use.
