The state of MLOps - machine learning in production at enterprise scale

Your ML code is just a small gear in a much bigger machine. Learn to build the robust systems required to run machine learning at enterprise scale.

#1about 4 minutes

The challenge of moving machine learning to production

While notebooks are great for experimentation, the actual machine learning code is only a small part of a much larger production system.

#2about 7 minutes

Adopting an MLOps mindset and culture

MLOps extends DevOps principles by integrating data scientists, developers, and operations into a single team with shared responsibility for the entire application lifecycle.

#3about 3 minutes

Best practices for managing production models

Treat models as versioned artifacts, continuously monitor them for data drift, and track business metrics to ensure ongoing accuracy and value.

#4about 8 minutes

A reference architecture for MLOps pipelines

A common MLOps pattern involves a batch pipeline for model training and a real-time pipeline for model scoring, a structure used by major cloud providers.

#5about 3 minutes

Navigating the complex MLOps tooling landscape

Choosing the right tools requires making foundational decisions between buying versus building solutions and using AutoML versus custom-built models.

#6about 7 minutes

Exploring different patterns for model serving

Models can be served in various ways, from monolithic application deployments to decoupled microservices or embedded directly within streaming applications for low latency.

#7about 1 minute

Containerizing ML applications for consistency

Using containers like Docker and orchestration platforms like Kubernetes provides a consistent, portable, and scalable environment for deploying machine learning applications.

#8about 3 minutes

The role and benefits of a feature store

A feature store is a specialized database that centralizes feature management, serving data efficiently for both batch model training and real-time model scoring.

#9about 2 minutes

Orchestrating MLOps workflows for reliability

Workflow orchestration tools like Kubeflow and Airflow are essential for scheduling, managing, and monitoring the batch jobs involved in training models and engineering features.

#10about 8 minutes

Further resources and Q&A on MLOps careers

A review of recommended books and courses for learning MLOps, followed by career advice on getting started in the field and the pros and cons of freelancing.