AI company Saltlux is pushing ahead with an "AI-based intelligent records information search solution development and demonstration project" led by the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA), it said on Wednesday.

The project is part of an inter-ministerial AI expansion initiative. It aims to improve search convenience and information use by applying AI to National Archives records.

Saltlux has so far indexed about 23.8 million record lists and about 540,000 original-text data fragments. It combines its in-house AI model LUXIA with an ontology-based knowledge graph and AI agent technology to analyze relationships between institutions, policies and tasks and the records.

For example, it can find related records even when an institution's name has changed or when the same policy is recorded under different names. The company plans to reduce hallucinations from generative AI by designing the system to produce answers based on the original text of actual records.

Saltlux is conducting integrated verification of data construction and the AI search function. It plans to pursue testing and certification and on-site demonstrations at the National Archives this month and improve search accuracy by reflecting user feedback.

Lee Kyung-il (이경일), chief executive of Saltlux, said, "Based on ontology and AI agent technology, we will create an environment where AI understands meaning and relationships within vast records, and where any member of the public can use the information they need together with grounds they can trust."