Blab AI
University of Maryland-Politecnico di Milano Team Wins NEA AI Coding Competition
The OECD Nuclear Energy Agency (NEA) announced that Syrra Team 1, composed of students from the University of Maryland and Politecnico di Milano, won its first coding competition. The contest, which attracted 40 teams worldwide, challenged participants to transform messy, unstructured construction risk registers into machine‑readable data.
The competition was held in the run‑up to the NEA International Workshop on Artificial Intelligence for Nuclear Energy in Jeju, South Korea. The winning team will travel to Jeju to present its methodology and results to NEA officials and industry stakeholders.
NEA’s objective was to address a practical problem that affects nuclear construction projects. Risk registers are widely used to document potential hazards, but they are often created in heterogeneous formats and contain free‑text descriptions that are difficult for computers to interpret. Converting these registers into structured data would enable automated analysis, improve decision‑making, and support the sector’s goal of expanding nuclear capacity.
To solve the problem, participants were asked to develop natural‑language‑processing pipelines that could read diverse risk register documents and output standardized, structured records. The competition emphasized the use of large language models (LLMs) and prompt engineering, but teams were free to combine other techniques as needed.
Syrra Team 1’s approach combined a state‑of‑the‑art LLM with carefully designed prompts and a custom post‑processing step that aligned the model’s output with a gold‑standard dataset created by nuclear risk‑management experts. According to the NEA, the team’s solution achieved the highest accuracy among all submissions, producing structured data that closely matched the expert‑generated reference.
The NEA’s own website notes that the competition “provided an opportunity to engage with and connect an exciting wave of young AI talent to the nuclear energy sector.” The agency’s mission is to assist member countries in maintaining and further developing the scientific, technological, and legal bases required for the safe, environmentally friendly, and economical use of nuclear energy. By involving students in a real‑world problem, the NEA hopes to accelerate the adoption of AI tools in nuclear construction.
The competition also highlighted the broader trend of applying AI to infrastructure projects. Similar initiatives have been launched in other high‑risk sectors, such as civil engineering and aviation, where unstructured data poses a barrier to automation. The NEA’s focus on risk registers reflects the sector’s priority of managing construction risks, a key factor in the cost and schedule performance of nuclear plants.
The NEA International Workshop in Jeju will bring together researchers, industry representatives, and policymakers to discuss the role of AI in nuclear energy. Syrra Team 1’s presentation will showcase the technical details of their pipeline and demonstrate how LLM‑based solutions can be integrated into existing risk‑management workflows.
At the time of writing, the NEA has not released a schedule for the workshop, but the agency has confirmed that the event will take place in late May 2026. The agency’s website lists the workshop as a key component of its broader AI‑for‑nuclear‑energy strategy, which includes research grants, data‑sharing initiatives, and cross‑border collaboration.
In summary, the NEA coding competition has identified a promising AI‑based method for converting unstructured risk registers into structured data. Syrra Team 1’s success illustrates the potential for LLMs and prompt engineering to address practical challenges in nuclear construction. The upcoming workshop in Jeju will provide a platform for the team to share its work and for the nuclear community to explore how AI can support safer, more efficient plant development.
The NEA’s initiative is part of a growing effort to integrate advanced analytics into the nuclear sector, a move that could help meet the industry’s long‑term goal of tripling nuclear capacity by 2050.
The competition was held in the run‑up to the NEA International Workshop on Artificial Intelligence for Nuclear Energy in Jeju, South Korea. The winning team will travel to Jeju to present its methodology and results to NEA officials and industry stakeholders.
NEA’s objective was to address a practical problem that affects nuclear construction projects. Risk registers are widely used to document potential hazards, but they are often created in heterogeneous formats and contain free‑text descriptions that are difficult for computers to interpret. Converting these registers into structured data would enable automated analysis, improve decision‑making, and support the sector’s goal of expanding nuclear capacity.
To solve the problem, participants were asked to develop natural‑language‑processing pipelines that could read diverse risk register documents and output standardized, structured records. The competition emphasized the use of large language models (LLMs) and prompt engineering, but teams were free to combine other techniques as needed.
Syrra Team 1’s approach combined a state‑of‑the‑art LLM with carefully designed prompts and a custom post‑processing step that aligned the model’s output with a gold‑standard dataset created by nuclear risk‑management experts. According to the NEA, the team’s solution achieved the highest accuracy among all submissions, producing structured data that closely matched the expert‑generated reference.
The NEA’s own website notes that the competition “provided an opportunity to engage with and connect an exciting wave of young AI talent to the nuclear energy sector.” The agency’s mission is to assist member countries in maintaining and further developing the scientific, technological, and legal bases required for the safe, environmentally friendly, and economical use of nuclear energy. By involving students in a real‑world problem, the NEA hopes to accelerate the adoption of AI tools in nuclear construction.
The competition also highlighted the broader trend of applying AI to infrastructure projects. Similar initiatives have been launched in other high‑risk sectors, such as civil engineering and aviation, where unstructured data poses a barrier to automation. The NEA’s focus on risk registers reflects the sector’s priority of managing construction risks, a key factor in the cost and schedule performance of nuclear plants.
The NEA International Workshop in Jeju will bring together researchers, industry representatives, and policymakers to discuss the role of AI in nuclear energy. Syrra Team 1’s presentation will showcase the technical details of their pipeline and demonstrate how LLM‑based solutions can be integrated into existing risk‑management workflows.
At the time of writing, the NEA has not released a schedule for the workshop, but the agency has confirmed that the event will take place in late May 2026. The agency’s website lists the workshop as a key component of its broader AI‑for‑nuclear‑energy strategy, which includes research grants, data‑sharing initiatives, and cross‑border collaboration.
In summary, the NEA coding competition has identified a promising AI‑based method for converting unstructured risk registers into structured data. Syrra Team 1’s success illustrates the potential for LLMs and prompt engineering to address practical challenges in nuclear construction. The upcoming workshop in Jeju will provide a platform for the team to share its work and for the nuclear community to explore how AI can support safer, more efficient plant development.
The NEA’s initiative is part of a growing effort to integrate advanced analytics into the nuclear sector, a move that could help meet the industry’s long‑term goal of tripling nuclear capacity by 2050.
