Churn prediction from model to production

The brief

A SaaS company wants churn predictions inside its CRM every morning, not in a notebook. Build the churn model, then take it to production: serve it as an API in Docker, track versions in a model registry, test it automatically, and watch for drift as customer behaviour changes. Good looks like: a reproducible training pipeline, a deployed API, a canary release of a second model version, and a drift report that triggers retraining.

What you will build

1. Build and evaluate the churn model with a reproducible pipeline
2. Track experiments and register models in MLflow
3. Serve the model as a FastAPI service in Docker
4. Add CI with tests and a release gate on model quality
5. Release a new model version as a canary and roll back if needed
6. Monitor data drift and trigger retraining

Free, open data to build it with

• Telco Customer Churn ↗ Kaggle (IBM sample data)
• KKBox churn prediction challenge ↗ Kaggle

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.