Deploying machine learning models in a repeatable, auditable way is one of the hardest practical challenges in data science. This talk uses a concrete value estimation example — predicting property or asset values — to demonstrate an end-to-end MLOps workflow powered by MLflow. We'll cover: - Experiment tracking: logging parameters, metrics, and artefacts across model iterations - Model registry: versioning models and managing staging vs. production transitions - Reproducibility: packaging environments and pinning dependencies so results are consistent - Serving: exposing the best model as a REST endpoint using MLflow's built-in serving tools By the end of the session, attendees will understand how MLflow integrates into a typical scikit-learn or XGBoost pipeline, and how the same patterns extend to more complex architectures. All code examples are Python-first and beginner-friendly enough for data scientists who haven't yet adopted MLOps tooling.