1.3.1. SageMaker as the Central Hub

💡 First Principle: SageMaker centralizes ML workflows so that data, models, and metadata flow between stages without manual glue code. The cost of this centralization is learning its component boundaries—and the exam tests exactly those boundaries.

SageMaker Studio is the IDE where most interactions begin. Think of it as the "control tower" that gives you access to every SageMaker component through a unified interface. From Studio, you can launch Data Wrangler for feature engineering, kick off training jobs, configure endpoints, and review model monitoring results—all without leaving the environment.

The power of centralization shows up in lineage tracking. When SageMaker Pipelines orchestrates a workflow, every step—which dataset was used, which hyperparameters were set, which model artifact was produced—gets recorded automatically. This solves the reproducibility problem discussed in 1.1.1 and creates an audit trail that Domain 4 security questions test.

But centralization has trade-offs. SageMaker Studio runs in a specific AWS Region, and its compute costs can accumulate. For simple, occasional inference tasks, deploying a Lambda function with a lightweight model may be more cost-effective than maintaining a SageMaker endpoint. The exam tests your ability to recognize when SageMaker is the right tool and when it's overkill.

⚠️ Exam Trap: "Use SageMaker" is often a distractor when the scenario describes a simple, low-frequency task. If a question describes processing a few hundred records once a week, SageMaker Batch Transform might work—but a Lambda function with the model loaded from S3 could be cheaper and simpler. Look for volume, frequency, and cost constraints in the question.

Reflection Question: When would you choose AWS Glue over SageMaker Data Wrangler for data preparation, and what signals in an exam question would point you toward each?