Show HN: ML-Ralph – An autonomous agent loop for ML experimentation

We adapted RALPH for ML workflows.

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