Model documentation rots because it's manual. modelcard-gen makes it a build step: keep a small JSON config next to your model and generate a clean Model Card on every release.
pip install -e .modelcard-gen examples/fraud_scorer.json -o MODEL_CARD.mdfrom modelcardgen import render
print(render({"name": "Fraud Scorer", "metrics": {"roc_auc": 0.97}}))Scalar dicts (like metrics) render as tables; lists render as bullets. Sections follow the standard Model Card layout (details, intended use, data, metrics, limitations, ethics). MIT licensed.