Heart Disease MLOps — MLflow + Streamlit Deployment | Luis Núñez

What it covers

The full lifecycle of a small ML model — from data prep through production deployment:

Data prep — heart disease dataset (918 patients, 7 features), train/test split
Training with MLflow tracking — two parallel approaches (Random Forest, Dense NN) both logged to MLflow with parameters, metrics, and model artifacts
Model registry — register the chosen model and promote to the prod alias
Deployment in two patterns:
- Embedded: Streamlit loads the MLflow prod model directly into its process
- API-backed: FastAPI wraps the model; Streamlit calls it over HTTP

Embedded is simpler — one process, one deployment. Fine for demos, small-scale internal tools.
API-backed is how production ML systems are usually structured — model service can scale independently from the UI, can be called by multiple clients, and decouples the model release cycle from the app release cycle.

Showing both side-by-side demonstrates that model serving isn’t a single solved problem — it’s a design decision.