ml-intern-lab

An ML engineer agent sandbox for reading papers, running experiments, and shipping model reports.

This repo is inspired by the late-April 2026 trend around huggingface/ml-intern and agentic ML engineering workflows.

Goal

Create a small lab where an agent can turn an ML idea into a tracked experiment: paper notes, dataset assumptions, training command, metrics, and a final report.

First Workflow

paper or idea -> experiment plan -> run baseline -> collect metrics -> write report

Features To Build

• Paper summary template.
• Experiment plan schema.
• Dataset card generator.
• Baseline training runner.
• Report writer with metrics and next-step recommendations.

Repository Shape

experiments/
  0001-baseline/
templates/
  paper-note.md
  experiment-plan.json
  model-report.md
src/
  runner/
  reports/

Milestone 1

• Done: add experiment templates.
• Done: implement a tiny local majority-class baseline runner with no external dependencies.
• Done: generate metrics.json and model-report.md.
• Done: add one sample experiment using a toy dataset.

Run It

python -m ml_intern_lab.cli experiments/0001-baseline/experiment-plan.json

Test It

PYTHONPATH=src python -m pytest

Publish

This package is ready for GitLab Package Registry and PyPI releases. See RELEASE.md.

Trend Notes

• ML agents are moving from chat to execution.
• Reproducibility is the selling point: every report should link to config, data assumptions, and metrics.
• Start with classical ML baselines before adding GPUs or deep learning complexity.