ML Intern Lab: A Minimal Agentic Workflow for Reproducible Machine Learning Experiment Reports

Machine-learning teams increasingly use AI agents to read papers, propose experiments, run baselines, and draft reports. These workflows are useful, but they often blur together planning, execution, metrics, and narrative output in ways that are difficult to audit. This paper presents ML Intern Lab, a minimal agentic workflow for reproducible machine-learning experiment reports. The workflow turns an idea or paper note into an explicit experiment plan, runs a local baseline, writes machine-readable metrics, and generates a short model report. The reference implementation uses a zero-dependency Python runner and a tiny majority-class baseline experiment to demonstrate the pattern. The contribution is not a new model or benchmark. It is a compact reporting workflow that helps agentic ML assistants leave inspectable artifacts at each step, making their work easier to review, reproduce, and extend.

The manuscript is accompanied by the source manuscript, metadata, BibTeX file, renderer, and upload bundle.