It runs experiments autonomously, forming hypotheses, training models, evaluating results, iterating on evidence.
W&B integration for long-running jobs. Full audit trail.
Tested on Kaggle Higgs Boson, hit top 30 in a few hours.
Still early, lots to improve. Would love feedback.
github.com/pentoai/ml-ralph