GODML

Governed, Observable & Declarative Machine Learning Framework

Production-grade MLOps for teams that need traceability, compliance, and a verified supply chain — without the infrastructure overhead.

Quick start

pip install godml
godml init my-project
godml run -f godml.yml

That's it. No cloud account required for local training.

What is GODML?

GODML is a Python framework that wraps the full ML lifecycle — data prep, training, evaluation, monitoring, and deployment — behind a single declarative YAML config. Every run produces a signed, auditable artifact trail.

Raw data → Compliance check → Train → Evaluate → Registry → Deploy → Monitor
               (PII/GDPR)    (XGB/RF/LR)  (cross-val)  (MLflow)  (Docker)  (drift)

Why GODML over plain sklearn + MLflow?

Installation

Core (no optional deps)

pip install godml

With extras

pip install "godml[advisor]"   # LLM-powered recommendations (gpt4all)
pip install "godml[deep]"      # LSTM forecasting (tensorflow + keras)
pip install "godml[aws]"       # SageMaker deployment
pip install "godml[api]"       # REST inference server (fastapi + uvicorn)
pip install "godml[dev]"       # Full dev suite (tests, lint, coverage)

Configuration

A minimal godml.yml:

name: customer-churn
version: 1.0.0
provider: mlflow

dataset:
  uri: ./data/churn.csv
  hash: auto

model:
  type: xgboost
  hyperparameters:
    max_depth: 6
    learning_rate: 0.1
    n_estimators: 300

metrics:
  - name: auc
    threshold: 0.85
  - name: accuracy
    threshold: 0.80

governance:
  owner: ml-team@company.com
  tags:
    - compliance: gdpr
    - environment: production

deploy:
  realtime: true
  batch_output: ./outputs/predictions.csv

Run it:

godml run -f godml.yml

Notebook API

For interactive work in Jupyter:

from godml import GodmlNotebook

nb = GodmlNotebook()
nb.load_data("./data/churn.csv", target="churn")
nb.train_model("xgboost", {"max_depth": 6, "n_estimators": 300})
nb.evaluate(["auc", "accuracy", "f1"])
nb.save_model("churn_v1")

AI-powered advisor

from godml.notebook_api import advisor_full_report, tune_model

# Get model + metric recommendations for your dataset
report = advisor_full_report(df, target="churn")
print(report["recommended_models"])   # ['xgboost', 'random_forest']
print(report["data_quality"])         # quality score + issues

# Auto-tune with Optuna
result = tune_model(
    model_type="xgboost",
    X=X_train, y=y_train,
    max_trials=50,
    metric="auc",
)
print(f"Best AUC: {result['best_score']:.4f}")

Supported model types

Compliance

from godml.compliance_service import PciDssCompliance, GdprCompliance

compliance = PciDssCompliance()
clean_df = compliance.apply(df)          # masks PAN, CVV, account numbers

gdpr = GdprCompliance()
report = gdpr.apply(df)                  # anonymizes PII per GDPR rules

Built-in compliance modules: PCI-DSS, GDPR, HIPAA, SOX.
Custom rules: subclass BaseCompliance and implement apply(df).

Architecture

┌──────────────────────────────────────────────────────┐
│                    GODML Framework                   │
├────────────────┬─────────────┬───────────────────────┤
│  Interfaces    │  Notebook   │  CLI  │  REST API      │
├────────────────┴─────────────┴───────────────────────┤
│  Core Services                                       │
│  ┌───────────┐ ┌───────────┐ ┌──────────────────────┐│
│  │ Advisor   │ │ Config    │ │ Pipeline Engine      ││
│  └───────────┘ └───────────┘ └──────────────────────┘│
├──────────────────────────────────────────────────────┤
│  ML Services                                         │
│  ┌───────────┐ ┌───────────┐ ┌──────────────────────┐│
│  │ DataPrep  │ │ Model     │ │ Monitoring           ││
│  │ +PII scan │ │ Registry  │ │ +Drift detection     ││
│  └───────────┘ └───────────┘ └──────────────────────┘│
├──────────────────────────────────────────────────────┤
│  Providers:  MLflow │ SageMaker │ Docker │ Local      │
└──────────────────────────────────────────────────────┘

Supply chain & security

GODML ships with a SLSA Level 3 supply chain — every release is built in an isolated GitHub Actions environment with unforgeable provenance.

Verify the SBOM yourself

# Download from GitHub Releases
cosign verify-blob \
  --bundle sbom.spdx.bundle \
  --certificate-identity-regexp "https://github.com/DAGMALIA/godml/.github/workflows/safety_scan.yml" \
  --certificate-oidc-issuer "https://token.actions.githubusercontent.com" \
  sbom.spdx.json

Verify SLSA provenance

slsa-verifier verify-artifact dist/godml-*.whl \
  --provenance-path provenance.intoto.jsonl \
  --source-uri github.com/DAGMALIA/godml \
  --source-tag v1.1.0

CI security controls

CLI reference

godml init <project>         # scaffold new project
godml run -f godml.yml       # execute pipeline from config
godml deploy <project> <env> # deploy model to environment
godml --version              # print version

Roadmap

v1.2.0 — Q3 2026

• Interactive drift dashboard (Streamlit)
• A/B testing framework
• Optuna distributed tuning

v1.3.0 — Q4 2026

• Kubernetes operator
• Real-time streaming inference
• Multi-tenant model registry

v2.0.0 — 2027

• Multi-cloud provider abstraction (Vertex AI, Azure ML)
• Federated learning support
• SOC2 / ISO27001 documentation kit

Contributing

git clone https://github.com/DAGMALIA/godml.git
cd godml
pip install -e ".[dev]"
pytest tests/ --cov=godml

See CONTRIBUTING.md for branch conventions and PR checklist.

License

MIT — see LICENSE.

Built by DAGMALIA · PyPI · Support