Reinforcement Learning · Game AI
In production
Largest ongoing project
Geometry Dash Agent (vision-based)
A gameplay agent trained from pixels with a reproducible evaluation harness. The goal is strong generalization across levels through curriculum design and robust training pipelines.
PyTorch
RL (PPO-style)
CV pipeline
Evaluation
Training
Curriculum + reward shaping and stable rollouts.
Data
Support for learning from real levels (dataset ingestion).
Repro
Metrics, seeds, and sanity checks baked in.