An educational repository for teaching Machine Learning Operations (MLOps) and production ML engineering practices. This course covers the complete ML lifecycle from data processing through deployment and monitoring, with heavy emphasis on containerization, orchestration, and modern ML tooling.

The Playlist: https://www.youtube.com/playlist?list=PLOiItT5FLNRqN9DeGbnXpw2SMehnAXpax

Repository: https://github.com/wodecki/ASI_2025

Python Version: 3.10-3.11

Package Manager: uv (mandatory)

• Containerization: Docker, Kubernetes
• Cloud Platform: Google Cloud Platform (GCP)
• Data Processing: pandas, DuckDB, BigQuery
• ML Frameworks: MLFlow, AutoGluon, scikit-learn
• Web Frameworks: FastAPI, Streamlit
• Monitoring: MLFlow, WandB, Locust
• LLM Integration: OpenAI API

ASI_2025/
├── slides/                           # Course presentation materials
│   ├── 0. Introduction/
│   ├── 1. Set-up/
│   ├── 2. BigData/
│   ├── 3. Model Lifecycle/
│   ├── 4. Software Engineering/
│   ├── 5. ML Pipelines/
│   ├── 6. Containerization/
│   ├── 7. AutoML/
│   ├── 8. Experiment tracking/
│   ├── 9. Data and Model Drift/
│   ├── 10. Kubernetes and Kubeflow/
│   └── 11. AI Architecture Design/
└── src/                              # Hands-on code examples
    ├── 1. Set-up/
    └── 2. BigData/
        └── 1. Data/
            ├── 1. pandas_vs_duckdb/          # Performance comparison demo
            └── 2. BigQuery Iowa Transfer/     # GCP data pipeline tutorial
   └── ... to be continued ...

Introduction to MLOps concepts, course structure, and learning objectives.

Development environment configuration, Python setup, and tooling introduction.

• pandas vs DuckDB Performance Comparison - Benchmark data processing at scale (100M-1B records)
• BigQuery Data Transfer - Import public datasets to GCP infrastructure
• Efficient data processing patterns
• Cloud storage optimization

Key Demos:

• src/2. BigData/1. Data/1. pandas_vs_duckdb/ - Compare pandas and DuckDB on large datasets
• src/2. BigData/1. Data/2. BigQuery Iowa Transfer/ - Transfer Iowa Liquor Sales dataset from Google Public Datasets

CRISP-DM methodology, data understanding, preparation, and modeling workflows.

Best practices for production ML code: version control, testing, documentation, and code quality.

Building reproducible ML pipelines, data versioning, and workflow orchestration.

• Docker fundamentals
• Single and multi-container applications
• Volume management
• Production deployment patterns

Automated machine learning with PyCaret and AutoGluon for rapid prototyping.

Weights & Biases (WandB) integration for experiment management and model versioning.

Monitoring model performance, detecting drift, and triggering retraining pipelines.

Container orchestration, scaling ML workloads, and Kubeflow pipelines.

System design patterns, microservices architecture, and production ML systems.

• Python 3.10 or 3.11
• Docker Desktop (for containerization modules)
• Google Cloud Platform account (for cloud modules)
• Git

1. Clone the repository:

   git clone https://github.com/wodecki/ASI_2025.git
   cd ASI_2025

2. Install uv package manager:

   # macOS/Linux
   curl -LsSf https://astral.sh/uv/install.sh | sh
   
   # Or via pip (one-time)
   pip install uv

3. Navigate to a specific module:

   cd "src/2. BigData/1. Data/1. pandas_vs_duckdb"

4. Install dependencies:

   uv sync

5. Run examples:

   uv run python bigdata_pandas_vs_duckdb.py

Course Instructor: Andrzej Wodecki, [email protected]

Repository Issues: https://github.com/wodecki/ASI_2025/issues

This educational material is provided for learning purposes.

Last Updated: October 2025 Course Year: 2025