AI · 2026
AI Cup 2026
2nd of 89 teams in the Netherlands: bird species identification from radar data, for conservation around wind farms.
- Client
- Team Epoch (TU Delft), AIC4NL, TNO (competition)
- Built with
- PythonTabPFNLightGBMCatBoostXGBoost
Problem
Wind farms need to know which bird species are flying through, in real time, to reduce collisions. The AI Cup 2026, organised by Team Epoch at TU Delft with the AI Coalition for the Netherlands (AIC4NL), asked teams to identify species from TNO radar data and to propose how such a model fits an operational system.
Solution
Five weeks, two tracks. Team Geto-Dacians built the model for the Kaggle performance track: flight features with the wind’s effect taken out, gradient-boosted trees and TabPFN combined in an ensemble, and a correction based on real migration counts. Alongside it came the implementation proposal: system description, responsible-AI analysis, data collection and labelling, and mitigation measures.
Results
2nd place out of 89 teams and 300 students, in the largest student AI competition held in the Netherlands. Nominated to the final five and awarded 2nd place by the AIC4NL jury at the Dutch AI Congress on 14 April 2026, with a €2,000 prize.
