A data scientist has a churn model that works in a notebook. The retention team wants a score for every customer each morning, and they want to know when the model stops being right.
An ML or MLOps engineer takes models from a notebook to a service that stays correct. You package, deploy, monitor and retrain: versioned data and models, repeatable pipelines, and alerts when quality drifts.
Goal: Understand what machine learning is, train and evaluate a simple model in Python, and package it behind a small API.
Before you start
Beginner · 7 weeks
Week 1: What ML is
Learning from examples versus writing rules · Features, labels, training, prediction · Classification versus regression · Overfitting in one picture
Practice: Describe three everyday problems and say whether each is classification, regression or neither.
Practice: Describe three everyday problems and say whether each is classification, regression or neither.
Common pitfalls
Think first, then reveal
What is overfitting?
Think first, then reveal
Why do we split data into train and test?
Think first, then reveal
When is accuracy a poor metric?
Think first, then reveal
What is the difference between a feature and a label?
Training and serving use the same feature definitions, built from versioned data.
Code that builds model inputs from raw data, with the data snapshot recorded so a result can be reproduced.
Why this pick
Most production model bugs come from training features that differ from serving features.
What the role must know
Beginner mistake
ML
Python
Ops
Cost and reliability
Think first, then reveal
What is overfitting?
Think first, then reveal
Why do we split data into train and test?
Think first, then reveal
When is accuracy a poor metric?
Think first, then reveal
What is the difference between a feature and a label?
Think first, then reveal
What is train/serve skew and how do you prevent it?
Think first, then reveal
How do you detect model degradation when labels arrive late?
Think first, then reveal
What goes in a model card?
Think first, then reveal
How do you do a safe model release?
Think first, then reveal
How do you decide whether a problem needs ML at all?
Think first, then reveal
What are the common silent failures in production ML?
Think first, then reveal
How do you set SLOs for a model service?
Think first, then reveal
How do you reduce the cost of serving without losing quality?
Build one, measure it and put it on your CV. Each has tools, setup commands and milestones.