This repository provides materials for a CMU course on Machine Learning in Production (MLOps) and AI Engineering. It addresses building, deploying, assuring, and maintaining ML-powered systems, emphasizing responsible AI. Targeted at students with data science and programming skills, it aims to foster collaboration between data scientists and software engineers for robust AI system development.

How It Works

The course covers the end-to-end lifecycle of ML systems, from model development to production deployment. It focuses on designing resilient systems, ensuring safety, security, and fairness, and implementing robust deployment, testing, and monitoring. Key topics include MLOps automation, production experimentation, drift detection, scalability, and evaluating qualities beyond accuracy like latency and explainability.

Quick Start & Requirements

Publicly available course materials are on GitHub (https://github.com/ckaestne/seai/). Prerequisites include basic ML familiarity and essential programming skills (Python, libraries, command line). Students are expected to learn new tools independently. Slack access is recommended for in-class activities.

Highlighted Details

• Features a significant group project on building and deploying a movie recommendation service.
• Introduces production tools: Docker, Kafka, Jenkins, Prometheus, Grafana.
• Employs pass/fail grading with a token-based resubmission system.
• Focuses on interdisciplinary team collaboration.

Maintenance & Community

Materials are from CMU's Fall 2022 offering, led by Christian Kaestner. Communication channels include Slack, Canvas, and email. Materials are shared openly to encourage adoption at other universities.

Licensing & Compatibility

Course materials are shared under a Creative Commons license, with unspecified terms that may restrict commercial use.

Limitations & Caveats

The course is in-person only, with no online options or recordings. Students lacking basic ML and programming experience may struggle. The repository contains educational materials, not a deployable software product.