Skip to content

Domain 3 Review

Introduction

Original video transcript

This is the domain where many people feel least at home. Deployment sounds like someone else's job, a thing the DevOps team handles after the interesting work is done, right? And that assumption is exactly why models get stuck in notebooks. A model that never reaches production delivers no value, no matter how well it's scored. So let's look at what this domain actually is. Deployment and orchestration of ML and AI workflows is how a trained model becomes a running repeatable system, and it breaks into three tasks. Let's walk through them. Sure, the first task is managing deployment infrastructure. That means selecting compute environments and deployment targets, choosing among inference strategies like real-time and batch, and evaluating foundation model deployment options. The second task is provisioning and configuring resources. This covers building containers, configuring endpoints inside secure networks, setting up auto scaling, and deploying models and agents programmatically. And the third is automated orchestration and CICD pipelines for MLOps. You automate deployments and rollbacks, build retraining mechanisms, manage model versions, and refresh knowledge bases behind retrieval systems. 3 tasks and they stack. The infrastructure choices shape what you provision, and the pipelines automate both, so a deployment stops being a one-time event. So why does this matter for the machine learning engineer? Because getting models out of notebooks and into production reliably is the core of this role. Exactly. Training a good model is only the beginning. Your value shows when that model serves real traffic, survives updates, and redeploys without surprises. Build the deployment path well, and shipping a model becomes routine. Build it badly. And every release is a risky manual event. So here's how the review process works. Each task area gets a light pass over the AWS services that relate to it, and then a deeper review of the skills underneath. Some tasks also get a review together video where we work a scenario out loud, and the knowledge checks along the way tell you where to slow down. That's what's covered in this domain from trained model to running system in 3 connected pieces. Let's get started.

Domain 3 focuses on transforming a trained model from a notebook prototype into a reliable, repeatable, and automated production system. A model delivers business value only when it successfully reaches production and serves real traffic.

The 3 Core Tasks of Domain 3

  ┌─────────────────────────────────────────────────────────┐
  │ Task 1: Managing Deployment Infrastructure              │
  │ • Compute selection & deployment targets                │
  │ • Inference strategies (Real-time vs. Batch)            │
  │ • Foundation model deployment options                   │
  └────────────────────────────┬────────────────────────────┘
                               │
                               ▼
  ┌─────────────────────────────────────────────────────────┐
  │ Task 2: Provisioning & Configuring Resources            │
  │ • Container building                                    │
  │ • Secure network endpoints & Auto Scaling               │
  │ • Programmatic model & agent deployment                 │
  └────────────────────────────┬────────────────────────────┘
                               │
                               ▼
  ┌─────────────────────────────────────────────────────────┐
  │ Task 3: Automated Orchestration & CI/CD Pipelines       │
  │ • Automated deployments & rollbacks                     │
  │ • Automated retraining mechanisms                       │
  │ • Model versioning & Knowledge Base refreshes           │
  └─────────────────────────────────────────────────────────┘

1. Managing Deployment Infrastructure

  • Scope: Choosing the appropriate execution environments and deployment targets based on workload requirements.
  • Key Decisions: Selecting inference modes (e.g., synchronous real-time vs. offline batch transform) and evaluating hosting patterns for foundation models.

2. Provisioning and Configuring Resources

  • Scope: Turning infrastructure decisions into concrete, secure AWS resource configurations.
  • Key Capabilities: Authoring container images, configuring endpoints within secure private networks (VPCs), setting up auto-scaling rules, and programmatically deploying models and AI agents.

3. Automated Orchestration and CI/CD Pipelines for MLOps

  • Scope: Automating the end-to-end lifecycle so deployments are continuous rather than isolated manual events.
  • Key Capabilities: Implementing automated deployment and rollback strategies, setting up automated retraining loops, managing model artifact versioning, and refreshing Knowledge Base vector stores behind retrieval systems (RAG).

How the Tasks Stack Together

  • Infrastructure decisions dictate resource provisioning specifications.
  • CI/CD & orchestration pipelines automate both infrastructure and provisioning.
  • Impact: Moving from risky, manual releases to routine, automated production deployments that survive updates and redeploy without friction.