E2E
ML Lifecycle
FastAPI
Model Serving
Streamlit
Monitoring UI
RFM
Feature Engineering
Full ML lifecycle for customer churn prediction: feature engineering, model training with time-aware splitting, FastAPI model serving, and a Streamlit monitoring dashboard. Production patterns throughout — no notebook-only experiments.
The ML pipeline gap most tutorials skip: training-serving parity, time-aware splits, drift detection, production API.
Tech Stack
Pythonscikit-learnFastAPIStreamlitPandasDuckDBDockerPydantic
Features
Feature Engineering
RFM analysis, time-aware train/test splits, statistical feature selection.
Model Training
scikit-learn pipeline with hyperparameter tuning, cross-validation, SHAP explainability.
FastAPI Serving
REST API with Pydantic validation, async endpoints, health checks, versioned models.
Drift Detection
Streamlit dashboard monitors prediction distribution shifts and feature drift in production.
How It Works
01
Feature Store
Raw events → RFM features → DuckDB feature store
→
02
Train
Time-aware split → scikit-learn pipeline → SHAP explainability
→
03
Serve
FastAPI REST endpoint with Pydantic validation + versioned model registry
→
04
Monitor
Streamlit dashboard → drift detection → retraining trigger