AI Powers Nuclear Research with Secure Smart Tools
AI is reshaping nuclear research through secure, domain-specific tools that streamline simulations and data analysis for nuclear scientists,
How AI Is Quietly Transforming Nuclear Science
Graduate researcher Zavier Ndum is using text-based AI tools to automate and accelerate nuclear simulations—without compromising data security.
When people think of artificial intelligence, they might picture chatbots answering customer service questions or writing code on command. But AI, specifically large language models (LLMs), is making its way into more complex and traditionally high-stakes fields—like nuclear science.
At Texas A&M University, nuclear engineering Ph.D. student Zavier Ndum is leading efforts to harness AI in one of the most tightly regulated scientific domains. His pioneering research demonstrates how text-based AI tools can process nuclear data, run simulations, and even assist in public safety—all while safeguarding proprietary information.
Bringing ChatGPT-Like Tools into the Nuclear Lab
Ndum’s project, AutoFLUKA, acts like a virtual research assistant for nuclear engineers. Named after FLUKA, a particle transport simulation software, the tool can take input data, edit simulation parameters, run the models, and even generate graphical outputs. It’s powered by a secure LLM that never sends sensitive information off the user’s system—solving one of the biggest hurdles in AI adoption in this field.
“Nuclear science is built on data that isn’t always sharable,” Ndum explains. “You need systems that operate locally and securely, without exposing sensitive research or proprietary simulations to external servers.”
While LLMs like ChatGPT are impressive, they aren’t designed to handle confidential or domain-specific knowledge on their own. Ndum’s tool, on the other hand, allows researchers to create private, local databases and then ask the AI to extract or interpret data quickly and accurately—saving time and reducing errors.
A Timely Boost for Health Physics
Before diving into AI, Ndum’s expertise lay in health physics—the science of radiation protection. That background became a strength. He saw firsthand how time-consuming it could be for professionals like radiation safety officers to sift through lengthy documents and regulatory guidelines.
In one case study, Ndum tested his AI assistant to answer technical queries from health physics documentation. A search that would typically take hours took seconds. “Imagine needing to know the radiation dose limits for a specific machine and finding it instantly,” he said during a recent talk at the State of Texas Chapter of the Health Physics Society (STC-HPS).
This dual application—for both nuclear simulations and radiation safety—makes AutoFLUKA a versatile tool, with the potential to ease workloads across several specializations.
Bridging Two Worlds: AI and Nuclear Science
Building AutoFLUKA wasn’t straightforward. Ndum didn’t have access to Monte Carlo N-Particle (MCNP), a widely used nuclear simulation code that is more strictly regulated. Instead, he turned to FLUKA, which is similar in functionality but more accessible for academic research. The successful integration shows that his model could be adapted for other tools in the future.
His journey into AI was also a leap of faith. Originally focused on dosimetry—the measurement of radiation exposure—Ndum shifted gears with the encouragement of mentors like Dr. John Ford and Dr. Yang Liu, who saw the potential of AI to modernize nuclear research.
“It was challenging, stepping into unfamiliar territory,” Ndum admits. “But I believed there was value in connecting these fields—and the results speak for themselves.”
Expanding the Vision: Smarter AI, Smarter Research
Ndum isn’t stopping with simulations. He’s now developing an advanced LLM that can process multiple file types—PDFs, spreadsheets, images—and even pull relevant information from online databases in real time. This next-gen tool aims to answer complex, research-specific questions with nuance and accuracy.
Such capabilities could revolutionize nuclear science workflows, especially as the field faces a data boom from next-generation reactors and increasingly digital research environments.
Dr. Liu, who leads the Scientific Machine Learning for Advanced Reactor Technologies (SMART) group at Texas A&M, believes Ndum’s work is just the beginning. “Zavier’s approach represents a turning point in how we use AI in nuclear engineering,” Liu said. “By focusing on secure, domain-aware models, we’re not only improving efficiency—we’re protecting the integrity of the science.”
Conclusion: From Concept to Catalyst
Zavier Ndum’s story is more than a tale of technical innovation—it’s a blueprint for how AI can responsibly evolve into even the most sensitive scientific sectors. By aligning LLM capabilities with the unique needs of nuclear researchers and health physicists, his work is proving that smart tools don’t have to compromise on security to deliver value.
As research becomes more data-intensive and timelines grow tighter, tools like AutoFLUKA could become essential for scientists navigating both complexity and compliance. And with each breakthrough, the once-clear lines between AI and nuclear science blur a little more—for the better.
Disclaimer:
This article is intended for informational and educational purposes only. It highlights academic research and technological innovation without promoting any specific product or commercial interest. The views expressed are those of the individuals and sources cited.
source : phys.org