This repository provides a comprehensive course on MLOps, teaching how to build, deploy, and iterate on production-grade ML applications. It's designed for developers, data scientists, college graduates, and product managers seeking practical ML skills. The course emphasizes first principles, best practices, scalability, and CI/CD for robust ML systems.

How It Works

The course follows an iterative approach, starting from model experimentation and development, progressing to deployment and iteration. It integrates MLOps components like tracking, testing, serving, and orchestration. The core philosophy is to build a reliable production system by applying software engineering best practices and enabling seamless transitions from development to production without code or infrastructure changes.

Quick Start & Requirements

• Installation: Clone the repository, set up a Python 3.10 virtual environment, install dependencies (pip install -r requirements.txt), and install pre-commit hooks.
• Prerequisites: Python 3.10 recommended. Anyscale account is optional but recommended for a guided weekend learning experience with provided compute (GPUs). Local setup requires sufficient CPU/RAM.
• Resources: Detailed setup instructions for local machines and Anyscale workspaces are provided.
• Links: Course Lessons: https://madewithml.com/

Highlighted Details

• End-to-end MLOps workflow from experimentation to production.
• Integrates MLflow for experiment tracking.
• Demonstrates CI/CD pipelines using GitHub Actions for automated deployment.
• Supports local development and deployment on Anyscale, AWS, GCP, and Kubernetes.

Maintenance & Community

The project is maintained by Goku Mohandas. Community support and live cohorts are available via Anyscale.

Licensing & Compatibility

The repository content is licensed under the MIT License.

Limitations & Caveats

While the course covers various deployment targets, specific instructions and examples are heavily geared towards the Anyscale platform. Users deploying on other platforms may need to adapt configurations.