Atomic Canyon preps open-source nuclear search tool for release

An open-source model that “has set new benchmarks for accuracy, efficiency, and speed in AI search” will be available to ORNL, other national laboratories, independent researchers, and nuclear institutions. It will also be integrated into Neutron, an AI search platform that Atomic Canyon has developed.

Stats: After using ORNL’s Frontier supercomputer and sentence-embedding algorithms to train their AI model on the vast repository of licensing, regulatory, and technical documents in the Nuclear Regulatory Commission’s ADAMS document database, Atomic Canyon says the model is “capable of understanding complex nuclear terminology.”

The model “now returns the correct search result within the top ten results about 98 percent of the time, and within the top five results approximately 93 percent of the time, based on a new industry-first evaluation benchmark developed from real-world user queries,” the company said.

Its AI model “will help propel nuclear energy development,” said Thomas M. Evans, group lead and distinguished R&D staff member at ORNL. “What they have achieved is significant: It will be a valuable tool for the research, engineering, and deployment of nuclear energy and the transition to a future powered by sustainable energy.”

Atomic Canyon plans to continue its partnership with ORNL, using Frontier to refine and expand the model's capabilities over the next six months, with plans to release subsequent versions.

“The use of AI technology to do this sort of necessary but tedious work is a meaningful first step along the long road to licensing, and Frontier provides a terrific resource to train these types of models,” said Bronson Messer, director of science for the Oak Ridge Leadership Computing Facility, home to the Frontier supercomputer.

Improving data access: Atomic Canyon says its AI platform is designed for use in and by nuclear power plants, manufacturers of next-generation reactors, and government and national laboratories.

“Our collaboration with ORNL and the use of Frontier has enabled us to develop a tool that is not just a research milestone but a practical, commercial application that will transform data management within the nuclear sector,” said Kristian Kielhofner, cofounder and chief technology officer of Atomic Canyon.

The NRC has also been working on improving public access to the ADAMS document database.