★ Self-Retraining-Model-Platform ↗
MLOps platform running one closed loop. Drift detection triggers automatic retraining, conditional promotion, and zero redeploy serving.
More about this project
One FastAPI app over real infrastructure, Postgres, MinIO, MLflow, and Argo Workflows on Kubernetes. An 8 node DAG runs ingest, validate, feature engineering, Optuna search next to a PyTorch MLP branch, evaluation with a conditional promotion gate, then SHAP and promotion. A cron workflow batch scores every hour, and a React dashboard watches the loop live.
If I revisited it, I would use stronger drift statistics, MLflow model aliases instead of pinned stages, and deeper feature engineering.
- Python
- FastAPI
- Postgres
- MinIO
- MLflow
- Argo Workflows
- Kubernetes
- DVC
- Optuna
- PyTorch
- React
- Docker
- MLOps
- Drift Detection