Atomic Canyon AI Transforms Nuclear Data
The Nuclear Industrys Data Dilemma
Tech companies are betting heavily that nuclear power can supply the vast amounts of electricity required for their ambitious AI initiatives. However, data centers require this power imminently, and the nuclear sector is not typically characterized by rapid development.
Trey Lauderdale believes artificial intelligence can provide the nuclear industry with the necessary acceleration.
From Local Insight to Startup Vision
Lauderdale's fascination with nuclear energy began locally. Residing in San Luis Obispo, California, he frequently encountered employees from the Diablo Canyon Power Plant. He noted, "They’re like the coaches of our flag football team."
Through conversations with these individuals, he discovered that nuclear power plants are overwhelmed with documentation. For instance, the Diablo Canyon facility, located near his home, possesses approximately 2 billion pages of documents. Lauderdale, an experienced entrepreneur in the healthcare sector, suspected that AI could assist the nuclear industry in managing this extensive paperwork.
About a year and a half ago, Lauderdale founded Atomic Canyon, initially self-funding the venture. This startup leverages AI to aid engineers, maintenance technicians, and compliance officers in locating essential documents.
Securing Investment and Early Wins
Atomic Canyon secured a contract with Diablo Canyon in late 2024. According to Lauderdale, this agreement sparked interest from other nuclear power firms. "That’s when I knew, as an entrepreneur, we were at a point where we needed to raise a round of capital," he stated.
The company exclusively revealed to TechCrunch that Atomic Canyon has closed a $7 million seed funding round, spearheaded by Energy Impact Partners. Other participating investors include Commonweal Ventures, Plug and Play Ventures, Tower Research Ventures, Wischoff Ventures, and existing angel investors.
Tackling AI Challenges in a Specialized Field
Initially, Atomic Canyon’s AI engineers experimented with different models, but the outcomes were unsatisfactory. "We quickly realized the AI hallucinates when it sees these nuclear words," Lauderdale explained. "It hasn’t seen enough examples of the acronyms."
However, developing a new AI model demands significant computing resources. Consequently, Lauderdale arranged a meeting with Oak Ridge National Laboratory, a facility known for its nuclear research and home to the world’s second-fastest supercomputer. The laboratory found the concept compelling and granted Atomic Canyon 20,000 GPU hours of computing time.
Atomic Canyon's models employ sentence embedding, a technique well-suited for document indexing. These models are designed to make a nuclear power plant's extensive documentation searchable through retrieval-augmented generation (RAG). RAG utilizes large language models (LLMs) to generate query responses, compelling them to reference specific documents to minimize inaccuracies or "hallucinations."
Current Focus on Search and Future Aspirations
Currently, Atomic Canyon is concentrating on document search, partly because the associated risks are lower.
"One of the reasons we’re starting generative work around the titles of documents is because getting that wrong might cause someone a little frustration. It doesn’t put anyone at risk at the plant," Lauderdale commented.
Lauderdale envisions that Atomic Canyon’s AI will eventually be capable of creating "a first round draft" of documents, including references. "You are always going to have a human in the loop here," he emphasized.
However, Lauderdale did not specify a timeline for this development. Search is "the foundational layer," he asserted. "You have to nail the search." Furthermore, considering the sheer volume of documents in the nuclear sector, "we have a long runway in search alone," he added.
