Module 6
Module 6 - MLOps: From Script to Monitored Service
You will be able to turn a trained model into a tested, tracked, containerized service with a database behind it, tested and built on every push to main, and watched for the day its market moves.
A trained model in a script is a result, not a product. Making it one means a project anyone can rebuild from its lock (the file that records every package it installs at one exact version), tests of the code and of the model's behavior, a record of which run produced which model, a registry that names the version in use, a network interface with a schema in front of it, a container that runs the same everywhere, a database that keeps every prediction, a pipeline that checks every change, and monitoring that says when the inputs no longer look like the training rows. The module builds all of it around one model, Lesson 3.2.3's price pipeline on the course's listings, restated in full in its first lesson, so nothing from another module has to be on your machine.
This is where the earlier modules' work leaves the script. Every model before this one ran in the shared (bootcamp) environment and answered only a script that trained or loaded it. Here the same pipeline gets its own environment, MLflow 3.17.1 records and registers it, FastAPI serves it, Docker runs it beside PostgreSQL, GitHub Actions tests and builds it on every push to main, and the service reports its own health while a drift check reads what it was sent.
Without it, every result before this module stays on the machine that made it. A model nobody can retrain the same way, a score nobody can trace to a run, a service that works on one laptop, and the failure this module ends on: a deployed model that keeps answering fast, valid and wrong long after its market moved, while every dashboard says healthy.
What you will build
One project, the capstone, at the end of the module.
Ship It and Watch It Rot (PAID, $59, or $29 at the half price), the bootcamp's final project: a price service you ship, then catch pricing wrong with every response a 200, then feed a moved market until your own monitoring finds what its health report cannot, and retrain it, putting the new model in service only if it prices fresh sales better than the one it replaces. It starts from an empty folder and its setup writes everything it needs, the data and every stage of the service included.
The sections
- Project Discipline and Testing - 3 lessons. From script to project with uv, pytest (the tool that finds and runs a project's tests), then testing a trained model's behavior.
- Experiment Tracking and Serving - 4 lessons. MLflow tracking, logging and loading models from the registry, FastAPI's first endpoint, then the prediction service behind a schema.
- Containers, Databases and Pipelines - 4 lessons. Docker images and containers, PostgreSQL behind the service, layered images with
docker compose, then GitHub Actions. - Monitoring and the Closing Loop - 2 lessons. Logging predictions and service health, then data drift and the retraining loop.
Prerequisites
Modules 1 to 3, and Module 3's pipelines above all, the ColumnTransformer and the Pipeline that fill, encode and fit in one object, with Module 0's terminal, git and GitHub key, which this module uses constantly and never re-teaches. The model itself is restated in Lesson 6.1.1, and every library arrives pinned in the project's own lock.
What the exam proves
PAID, $59, or $29 at the half price. That you can say why a container fails to serve, read a failed pipeline correctly, tell a refusal that protects the model from an error that needs fixing, and tell a real drift signal from noise. The pass covers the capstone and the exam with every attempt, for the life of the product.
Set up your machine first
Module 6 leaves (bootcamp) behind: every lesson works in one project, ~/dev/listings-service, with an environment of its own that Lesson 6.1.1 creates. Module 6's setup page checks uv, git and your GitHub key, installs Docker Engine if it is missing, and ends with a checklist that must pass before Lesson 6.1.1.