Introduction
Nuclear energy ranks among the world’s most regulated industries. AI and especially generative AI have created enough impact that thought leaders rank it among other transformative “general purpose technologies” such as electricity and the steam engine. Harnessing AI to reimagine nuclear operations across the industry means more carbon-free nuclear energy for electrical grids and data centers, which the International Energy Agency estimates demand to double by 2026. In September 2024, Westinghouse unveiled its HiVE™ AI system, powered by its fine-tuned bertha™ generative AI model, transforming how customers collaborate with Westinghouse.
Building a Better Data Management Solution
Westinghouse’s digital transformation started more than five years ago with a deep bench of data and nuclear experts and over 70 years’ worth of cleaned and contextualized industrial data unique to the nuclear world. However, the team needed to improve the company’s data infrastructure if it wanted to realize its AI ambitions. The existing on-premises analytics database lacked some important scalability features and options. Without a scalable cloud solution, the data team struggled with a lack of computing resources, an inability to rapidly experiment with massive amounts of data, and restrictions on safely sharing data across applications.
To build a world-class, nuclear-specific AI capability, Westinghouse needed a better solution. Westinghouse decided to build on the Databricks Data Intelligence Platform, a move that would prove crucial in its mission to drive innovation. The nuclear industry has always been deeply committed to safety and reducing risk, with every detail inspected and regulated. Managing and securing critical nuclear data is not negotiable. With this in mind, Westinghouse set out to design a data backbone that could host AI applications for some of the most trusted utilities in the world. Databricks was the ideal partner to help Westinghouse achieve this goal.
As Westinghouse set out to design a data backbone so secure and robust that it could host AI applications for some of the most trusted utilities in the world, it turned to Databricks. The Databricks team quickly became a “guiding light” for Westinghouse, providing crucial support as the Westinghouse infrastructure team took the lead in configuring our systems to meet the nuclear industry’s strict regulatory requirements. Westinghouse was able to leverage Databricks’ state-of-the-art governance with Unity Catalog. It was built according to best practices outlined in the Databricks AI Security Framework (DASF), complementing Microsoft’s robust security standards. These foundations bolster the credibility of Westinghouse’s data management practices and give its customers peace of mind, which is essential in an industry where trust and reliability are paramount.
When it came time to modernize how the data was organized, the Databricks professional services team delivered. Together, Westinghouse and Databricks created a scalable and multi-tiered analytics environment, complete with an ML Ops process that streamlines the entire machine learning lifecycle. This foundation also featured a robust prototyping environment, along with dedicated workspaces, for testing and deploying AI models, all backed by a secure and reliable data lakehouse architecture.
The new infrastructure immediately saved hundreds of hours annually for the Digital Optimization Services business unit and allowed the Westinghouse team to reinvest in their product lines to include AI for customer-facing applications and services.
To make this vision a reality, Westinghouse had to ensure that its data was properly prepared, managed, and governed. That is where Databricks’ powerful technologies, including Auto Loader, Photon engine, and Lakeflow Jobs, really shined. Then, when Westinghouse needed real-time insights into its data quality and pipeline performance, they tapped into features like Lakehouse Monitoring and Expectations. Now, with Unity Catalog (UC) governing its data, Westinghouse has complete visibility into its data’s journey, from source to destination. In the nuclear industry, everything revolves around safety and trust. As Westinghouse continues to develop pioneering new AI solutions, Databricks services reinforce the trust Westinghouse earns for managing data securely and reliably.
Accelerating AI in a Complex Industry
On September 4, 2024, Westinghouse introduced its HiVE™ nuclear specific AI system and its bertha™ generative AI model to the world. Not only has the Westinghouse team rapidly advanced its AI capabilities using the Databricks Data Intelligence Platform, but it can now create future AI products and solutions limited only by imagination.
To assist in developing bertha™, Westinghouse leveraged the Databricks Mosaic AI Agent Framework, to rapidly evaluate various foundational models and GenAI systems. Using Databricks Experiments and MLFlow, Westinghouse conducted rapid experimentation to determine the best models, while logging statistics to evaluate performance. This approach enabled Westinghouse to accelerate the development of its custom Generative AI solution.
Westinghouse can now leverage its advanced data infrastructure to create solutions across the nuclear industry. For example, large industrial facilities communicate and store enormous quantities of data. With an architecture built on Databricks, Westinghouse maintains an AI solution to extract, cleanse, and store machine data from over 200 nuclear facilities worldwide. Another example includes an AI application designed to process video data in real-time inside Pressure Water and Boiling Water Reactors with the potential to detect debris at least 90% better than manual inspections and save up to 25% on inspection costs.
Lastly, another great example includes leveraging the bertha™ generative AI model to generate licensing data and documentation dramatically faster. Traditionally, it can take months to manually compile new nuclear site licenses or environmental assessments. This is a crucial step in streamlining nuclear development.
The Databricks infrastructure has freed data and nuclear experts to focus on nuclear innovation. As a result, the Westinghouse data scientists delivered four proofs of concept in December 2024, two production-grade systems in the first quarter of 2025, and helped generate 45 distinct innovation ideas in the first two months of 2025.
Conclusion
The Westinghouse-configured Databricks Data Intelligence Platform removes huge barriers to achieving Westinghouse’s AI ambitions. Now, Westinghouse can scale compute, rapidly and safely experiment with mass amounts of production data, and share information securely across applications. Westinghouse HiVE™ nuclear-specific AI system customers appreciate the power of auditability, input and output transparency, real-time data processing, and operational analytics. The Westinghouse teams value the incredible and adaptable partnership with Databricks to create a unique platform that positions it for continued pioneering AI innovation.
“With Databricks always providing the latest features that hit the market, Westinghouse is able to continually incorporate new AI capabilities for our customers.”
— Catherine Stanley, Data, Digital, and AI Manager at Westinghouse
