Arcadia AI — Master's Program
EN
РусскийO'zbekchaEnglishDeutschSrpskiहिन्दी
Program  /  Semester 2 — Modern AI Stack
Course

MLOps: Data Pipelines, Feature Stores, Monitoring, A/B Testing

Production ML: data pipelines, reproducibility, deployment, and monitoring

Instructor to be announced

About the course

This course builds the engineering culture needed to run ML systems in production. Students work through the full cycle from raw data to a live service: ETL, feature stores, model registries, CI/CD for ML, A/B testing, and data-drift monitoring. The course ties in closely with Practicum 2 and provides the toolkit required for the team industrial project in semester 3.

What you'll learn

✓Build a reproducible ML pipeline from raw data to a deployed model
✓Set up a feature store that guarantees training/serving parity
✓Implement CI/CD for an ML project with automated data and model validation
✓Run an A/B test for an ML component with sound statistical design
✓Set up drift monitoring and automated alerts for model degradation

Key topics

•Data pipelines: Airflow, dbt, batch vs streaming (Kafka, Flink)
•Feature stores: offline/online split, time-travel, backfilling
•Experiment tracking: MLflow, W&B; model registry and versioning
•Training/serving parity and distribution shift
•Containerization and orchestration: Docker, Kubernetes for ML
•CI/CD for ML: model testing, data schema validation, shadow deployment
•A/B testing and canary releases
•Monitoring data drift and quality degradation in production
•Cost optimization: spot instances, inference caching, batching
This description was generated automatically and has not yet been reviewed by an instructor — it's a draft for discussion.