This repository documents my journey participating in the ARC Prize 2026 on Kaggle.
The Abstraction and Reasoning Corpus (ARC) is a benchmark designed to evaluate an AI system’s ability to reason and generalize to unseen problems, not just memorize patterns. (Epoch AI)
- Learn reasoning-based AI
- Build rule-based and program synthesis solutions
- Improve generalization ability
- Joined ARC Prize 2026
- Built baseline model (zero output)
- Submitted first solution
- Python
- NumPy
- Solve ARC tasks manually
- Build rule-based solver
- Explore program synthesis approach
ARC is not about deep learning scale, but about discovering rules from small examples and applying them to new problems. (GitHub)