How mubashir251-tech/advance-data-science-secure-platform works
This repository is a runnable secure-data-science lab, organized around a loan-model lifecycle and six demonstration areas. Its principal workflow schedules training, fairness review, and a compliance policy gate; separate demos cover privacy-preserving learning, adversarial defenses, explainability, and auditability. The FastAPI app is a standalone demonstration API: its prediction endpoint uses a hard-coded rule, not the trained model. The README describes DevSecOps and platform services, but their implementation is not present in the supplied tree, so they are not mapped as repository components.
Main entry points: secure_training_dag.py Training DAG · 01_train_with_mlflow.py Model training · 11_policy_check.py Policy check · 02_bias_audit.py Bias audit
Components 19
MLOps lifecycle
- Training DAG
pipelines/dags/secure_training_dag.py - Model training
pipelines/01_train_with_mlflow.py
Governance
- Bias audit
governance/02_bias_audit.py - Explainability
governance/03_explainability.py
Privacy learning
- Anonymization
privacy/04_anonymize.py - Differential privacy
privacy/05_differential_privacy.py - DP-SGD training
privacy/06_dp_sgd_training.py - Federated learning
privacy/07_federated_demo.py
Model defenses
- Evasion attacks
adversarial/08_evasion_attack.py - Poison detection
adversarial/09_poisoning_detect.py - Robust training
adversarial/10_robust_training.py
Compliance
- Policy check
compliance/11_policy_check.py - Audit trail
compliance/12_audit_trail.py - Standards mapping
compliance/13_rmf_mapping.py
Other components
- Lab operator
- Loan dataset
- MLflow registry
- OPA service
- FastAPI service
fastapi/app.py
Connections 19
- Lab operator → Training DAG: triggers
- Training DAG → Model training: runs
- Training DAG → Bias audit: runs
- Training DAG → Policy check: runs
- Loan dataset → Model training: reads
- Model training → MLflow registry: logs model
- Loan dataset → Bias audit: reads
- Loan dataset → Explainability: reads
- Loan dataset → Anonymization: reads
- Loan dataset → Differential privacy: reads
- Loan dataset → DP-SGD training: reads
- Loan dataset → Federated learning: reads
- Loan dataset → Evasion attacks: reads
- Loan dataset → Poison detection: reads
- Loan dataset → Robust training: reads
- Policy check → OPA service: checks policy
- Lab operator → Audit trail: runs
- Lab operator → Standards mapping: runs
- Lab operator → FastAPI service: requests