The NEA is launching a coding competion. The competition's objective is for participating teams to create machine readable risk registers from diverse human readable risk registers using natural language processing and large language models (LLMs). 

Goal

Participants will create an algorithm or model that ingests diverse risk registers in Excel, PDF and Word formats, and produces standardised machine-readable risk registers for building machine learning models.

Prize

The Agency will provide travel expenses up to a total of €2 500* for the first-place team to present their team’s work at the upcoming International Workshop on Artificial Intelligence for Nuclear Energy. This event will take place in Korea in May 2026. The prize may be applied to one representative from the winning team or shared among team members.

*Subject to OECD Financial Regulations and other relevant OECD rules.

Participating teams will be provided with five rough input risk registers and three corrected output risk registers. Two risk registers will be selected as blind tests. Participants will build an algorithm or model that includes natural language processing and LLMs to convert the rough risk registers into machine readable corrected risk registers.  

The team is allowed to use the three corrected risk registers on which to train and test their algorithm or model. When the team feels their algorithm is adequate, they will run it on the two blind risk registers and produce Excel sheets in the format of the corrected risk registers. Teams will submit the resulting corrected risk registers by email, together with their model, for evaluation.

• Team size is limited to a maximum of four people. 

• Participation is limited to university students currently enrolled in an undergraduate or graduate degree seeking program (or graduated in the past 12 months).

• Only one final submission will be accepted per team.  

• The deadline for final submissions is 27 March 2026.  

• Submissions must include a pledge that the rules have been adhered to. 

The prize for winning will cover registration fees, travel costs, lodging and food up to a total, but not exceeding €2 500. This ammount which may be used for one representative from the winning team or shared among multiple team members, to be determined by the winning team.

Grading will be twofold: 

• First, a comparison algorithm will be used to compare the output files to the known corrected risk registers. Any difference between the two on the mandatory fields will subtract points. The score will be scaled to 100 and will account for 60% of the final weighted grade of the team.  

• Second, experts will compare and rank all received submissions in a blind ranking process to compare which submissions are most like the final corrected risk registers using human judgement. The score will be scaled to 100 and will account for 40% of the final weighted grade of the team. 

The team with the highest combined score will be declared the winner. In the event of a tie, the team with the highest automated score (first criterion) will be awarded first place. If a tie remains, the team that submitted their solution first will be awarded first place.

To register, teams are encouraged to email Jordan Cox (jordan.cox@oecd-nea.org) before 27 February 2026 with the following information: 

• Team name 

• List of team members with their email addresses

• Associated university 

Unprofessional team names and emails without an academic affiliation will be disqualified. 

Eligibility is restricted to nationals from NEA member countries, OECD member countries, OECD accession countries, Generation IV International Forum (GIF) member countries, and International Framework for Nuclear Energy Cooperation (IFNEC) participating and observer countries. Exceptions may be granted on a case-by-case basis. Please feel free to contact us to confirm your eligibility. 

Submissions should be via email and must include a zip folder containing model.py and the final two excel sheets output by model.py. Excel sheets should follow the naming convention of all other provided excel sheets. The code must be written in Python. The model.py should be a self-contained file that includes everything but an API key to call the LLM, which will be provided seperately. It should point to a directory for input files "input/" and output files to an output "output/"directory. The model should produce one output per input file. Submission deadline is 27 March 2026 at 24:00 CET.