AI agent that spots research flaws before work begins takes top prize at R&D hackathon

An AI agent that scans a research proposal for flaws before a single experiment begins, and another that instantly checks whether an expense receipt complies with funding regulations — ideas like these have moved from concept to working prototype.

The National Research Council of Science and Technology's National AI Research Center held the finals of the 2026 NAIS AI Hackathon at Seoul Dragon City from Wednesday to Thursday, selecting four winning teams.

Sponsored by the Ministry of Science and ICT, the competition centered on the theme "R&D-specialized AI agents: planting AI agents in every moment of research." A total of 160 teams applied, and 10 advanced past the preliminary round to compete in the finals.

Participating teams developed AI agents designed to tackle real-world problems across the full research cycle — from planning and literature review to experiments, data analysis, paper writing and administrative tasks. Each team refined its concept through expert mentoring and intensive development sessions, ultimately delivering a functional prototype.

The grand prize, awarded in the name of the deputy prime minister and Minister of Science and ICT, went to Team Daeway for "Project Neumann: Research Pre-mortem Agent Based on Evaluation History."

The agent cross-references publicly available peer-review and rejection records against a research proposal to identify structural flaws and risk factors before work begins. It generates "risk cards" citing specific evidence for each problem and suggests revisions. Judges praised the agent's technical completeness and its potential for direct application in real research settings.

The excellence award went to Team AI Strategy for an agent that identifies emerging research topics with strong growth potential. Team Geobugijeongshin took the merit award for an on-device research administration agent that checks expense receipts against relevant regulations at the moment of payment.

The innovation award went to Team m-M-m (media-Mediate-medicine) for an AI agent supporting research ethics administration. The system traces the legal basis of relevant provisions to map out review pathways, and is designed to explicitly flag areas where AI judgment remains unreliable.

The competition focused on moving beyond generative AI that simply answers questions, toward "agentic AI" capable of making independent judgments and carrying out tasks in pursuit of a given goal — applied directly to real R&D environments. The National Research Council of Science and Technology said it plans to review how winning entries and other standout ideas could be deployed in research settings and connected to follow-on studies.

Yoo Yong-gyun, director of the National AI Research Center, said the organization would continue working to deploy and refine the ideas that emerged from the competition "so they can grow into new colleagues that expand what researchers can ask and achieve."

NST Chairman Kim Young-sik said the event confirmed that creative ideas from the public, combined with the research capabilities of NST-affiliated government-funded institutes, could translate into genuine research innovation. "We will actively support the best outcomes so they can be applied and spread throughout real research environments," he said.