ml-experiment-runner

Run ML/deep learning experiments on Modal with robust error handling, W&B logging, and thorough documentation. Use when asked to: (1) Run an experiment, (2) Implement experiment handlers, (3) Fix and re-run failed sub-experiments, (4) Update findings after experiments complete. Emphasizes showing work, error recovery, and acting as a collaborative research colleague. Works with research_plan.yaml → handlers → results.yaml → FINDINGS.md workflow.