A database technology has been developed that can reduce "hallucination," cited as the biggest problem when enterprises adopt artificial intelligence (AI) agents.
A research team led by Professor Kim Min-soo of the School of Computing at the Korea Advanced Institute of Science and Technology (KAIST) said Wednesday it has developed a next-generation database technology called "AkasicDB" and a new retrieval-augmented generation (RAG) technique called "Omni RAG" that uses it. Using the new technology, AI answer accuracy improved by up to 78 percent and processing speed improved by up to 20-fold, the team found.
AkasicDB is an integrated database that can analyze multiple types of data at once. Typically, enterprise internal data is stored in various forms, including documents such as reports and contracts, table-format data, and relationship information between people, companies, and products. However, existing AI had to search this information separately, often failing to find sufficient supporting evidence or producing incorrect answers.
In response, the research team integrated a vector database that finds the meaning of documents, a graph database that analyzes relationships between entities, and a relational database that manages table-format data into a single system. This allows AI to simultaneously understand document content, relationships between data, and structured information.
For example, a query such as "Find clauses related to Company A among contracts signed last year, and explain how those clauses connect to product supply issues" previously required document search, relationship analysis, and date-condition search across multiple databases, followed by combining the results. This caused management complexity and response delays.
Omni RAG can process this at once. By integrating vector similarity search, graph traversal, and relational filtering within a single query and execution plan, and utilizing the information simultaneously, it finds more accurate supporting evidence and greatly reduces AI hallucination.
AkasicDB, which supports this, is also designed to perform multiple search methods in a single query. It is characterized by improved processing efficiency through minimizing the generation and movement of intermediate data.
In experiments, AkasicDB processed complex search queries—which took up to 21.3 seconds in existing systems—in under one second, achieving a performance improvement of more than 20-fold. Omni RAG was also found to improve answer accuracy by up to 78 percent compared to existing RAG.
The research team expects this technology to ease the hallucination problem, a key challenge for enterprise AI agents, and to enhance the practicality of AI services that use enterprise internal data.
"For AI agents to accurately understand and utilize a company's vast data, a data infrastructure that can integrate and process vector, graph, and relational data in a single system is essential," Professor Kim Min-soo said. "AkasicDB is a next-generation database technology for the era of AI agents, and we expect it to be used as core data infrastructure in fields requiring high reliability, such as defense, manufacturing, finance, law, and science and technology."
The research was conducted in collaboration with Graphai, a faculty startup, and the findings were presented as a demo paper at ACM SIGMOD 2026, the most authoritative international conference in the database field, on June 2.
