Home / Kaggle

Kaggle

I use Kaggle to practise machine learning against hard deadlines and honest scoring. Here's where each competition stands and what part of the score is my own work.

Kaggle profile
CompetitionPositionMy contributionEnds
Name That LLM436 of 4,346 (top 10%)Extra data pool, LoRA fine-tunes, n-gram SVM tuning, leak-free validationJan 2027
Mallorn ChallengeTop 15%Gradient-boosted models (XGBoost, CatBoost, LightGBM) tuned with OptunaEnded
Squares in Squares190 of 1,473 (top 13%)Validator and submission pipeline; the score comes from combining public packingsJul 2027
NeurIPS 2026 Smart BuildingsScore 0.01244Own physics-informed controller and unseen-building validation; found and reported a simulator bugJan 2027
Enveda CASMI 2026MRR 0.415Leak-free 300-structure hold-out and own scorers; the score comes from a credited public forkDec 2026
CAFA protein functionNot reportedResNet multi-label model for GO terms with extreme class imbalanceEnded

How I report ranks. Positions are public-leaderboard figures from 2 November 2026 and can change when each private leaderboard is published. I never claim a score that came from someone else's public notebook, and I say so where a score comes from a fork.