The Frontier supercomputer at Oak Ridge National Laboratory (ORNL) is opening a new chapter for artificial intelligence in the nuclear industry by enabling the creation of AI models trained specifically on nuclear data. Frontier, one of the most powerful supercomputers in the world, was used by tech start-up Atomic Canyon and its partner, the Diablo Canyon nuclear power plant in California, to build AI tools that can sift through massive archives of nuclear documents, regulations, and maintenance records much faster and more accurately than traditional methods, tells Tech Xplore.
The nuclear sector produces and depends on millions of pages of highly technical documentation, from engineering evaluations and safety procedures to regulatory filings. Diablo Canyon alone has about 2 billion pages of documents, and plant engineers often spend thousands of hours each year searching them for specific information needed to support licensing, inspections, and troubleshooting. Existing commercial AI models struggle with nuclear terminology and context, so the team built bespoke models that understand the industry’s language and intricacies.
Using Frontier’s massive GPU resources, Atomic Canyon developed the Neutron platform, which includes advanced sentence-embedding AI models nicknamed FERMI models. These models assign numerical representations to nuclear vocabulary and context, allowing users to retrieve relevant documents quickly, even across disparate databases and the Nuclear Regulatory Commission’s expansive archives. Training these models required Frontier’s high-performance computing power to process millions of pages repeatedly until accuracy and reliability were achieved.
Early results from Diablo Canyon show a tangible lift in productivity. Engineers can now extract detailed histories of specific components or regulatory requirements in minutes rather than weeks, freeing them to focus on technical problem-solving instead of administrative search tasks. The approach reduces labor, accelerates licensing work, and improves overall data access.
Atomic Canyon plans to expand these AI tools and make foundational models open source so other nuclear facilities can benefit. ORNL researchers are also exploring ways to combine these retrieval models with generative AI for even broader applications, from automated reporting to advanced simulations tailored to nuclear operations.