Skip to main content
In this lesson we examine why it’s helpful to approach Amazon SageMaker by persona. Mapping SageMaker features to the activities different people perform in the ML lifecycle clarifies responsibilities and shows how each capability fits into real workflows: data preparation, training, deployment, and monitoring. We focus on three common personas—Data Engineer, Data Scientist, and MLOps Engineer—and explain which SageMaker tools each persona typically uses. SageMaker can be used via the point-and-click Console or via the SageMaker SDK for Python (a code-first workflow). For reproducibility, version control, and automation—critical in production—we strongly recommend the code-first approach (typically from a Jupyter or JupyterLab environment).
Prefer a code-first workflow using the SageMaker SDK from a notebook (Jupyter / JupyterLab) for reproducible, versioned, and automatable ML work. The Console is useful for exploration and quick experiments, but production-grade pipelines benefit from code, CI/CD, and infrastructure-as-code.
We summarize the three personas and their high-level responsibilities next:
A presentation slide titled "Personas – Introduction" listing three roles: Data Engineer, MLOps Engineer, and Data Scientist. Each role has bullet points summarizing responsibilities (data warehousing/ETL, ML pipelines/CI‑CD/versioning, and experimentation/feature engineering/training/inference).
PersonaPrimary FocusTypical Responsibilities
Data EngineerData ingestion, transformation, governanceBuild repeatable ETL/ELT pipelines, ensure PII / encryption / access controls, stage data in S3 or data lakes
Data ScientistExperimentation and model developmentFeature engineering, training experiments, hyperparameter tuning, notebook-driven workflow
MLOps EngineerProductionization, automation, and monitoringCI/CD for models, model registry, deployment, autoscaling, drift detection, governance and traceability
Now we’ll examine each persona in detail and list the SageMaker capabilities they commonly use.

​
Data Engineer

Data Engineers locate, ingest, and transform source data so it becomes reliable training data. Sources include relational databases (MySQL, PostgreSQL, SQL Server, Oracle) and non-relational stores (DynamoDB, MongoDB, Redis). Ingestion can be ad hoc exports for experimentation or fully automated pipelines for production retraining. Transformations often required before training:
  • Select relevant columns and types.
  • Aggregate records or compute rolling statistics.
  • Remove or obfuscate personally identifiable information (PII).
  • Reformat to efficient columnar formats (Parquet) for large-scale training.
Amazon S3 is the common staging area for semi-structured datasets (CSV, Parquet, JSON). Data Engineers must apply governance (PII handling, encryption, IAM controls) before exposing datasets for model training or feature production.
A dark presentation slide titled "Data Engineer" with an icon of a table and padlock and the text: "Governance/Privacy constraints may require obfuscating or dropping parts of the source data." A small "© Copyright KodeKloud" appears in the bottom corner.
In smaller teams, Data Scientists may perform extraction and preprocessing manually from notebooks. In larger organizations, Data Engineers design automated pipelines (Python scripts, orchestration tools, or AWS services like AWS Glue) to ensure fresh, consistent, and governed datasets are available for training and inference.
A slide titled "Data Engineer" comparing small organizations (where a data scientist handles data extraction and transformation manually) with large enterprises (where a data engineer builds an automated pipeline to ingest and transform training data).
Key SageMaker tools for Data Engineers:
  • Data Wrangler: low-code visual transformations and repeatable ETL-like flows.
  • SageMaker Processing: run scalable Spark or Python processing jobs outside notebooks.
  • SageMaker Feature Store: persist and serve engineered features to ensure consistency across training and inference.
  • SageMaker Pipelines: orchestrate extract/transform/load and handoffs into training/validation stages.
A slide titled "Data Engineer" showing four numbered SageMaker components—01 Data Wrangler, 02 SageMaker Processing, 03 SageMaker Feature Store, and 04 SageMaker Pipelines—with brief descriptions of each.

​
MLOps Engineer

MLOps Engineers focus on safely getting models from experiments into production and keeping them reliable, performant, and compliant. Their responsibilities span deployment, autoscaling, lifecycle automation, monitoring, and governance. Core MLOps responsibilities:
  • Design CI/CD pipelines that test code and data, run training, register models, and gate deployments.
  • Manage a model registry for versioning artifacts and controlling approvals.
  • Deploy models (SageMaker Endpoints or other hosting) with autoscaling, A/B/blue-green strategies, and rollback mechanisms.
  • Monitor production models for performance drift, data drift, latency, fairness, and explainability; trigger retraining when necessary.
  • Maintain lineage and traceability: which code, dataset, algorithm, and model created a prediction.
A slide titled "MLOps Engineer" that lists key governance responsibilities. The four boxes say: enforces governance policies across the ML pipeline; ensures traceability of models, datasets, and code versions; automates compliance checks in CI/CD; and monitors deployed models for drift, fairness, and explainability.
MLOps best practices mirror DevOps: keep source code in Git, trigger pipelines on commits (linting, unit tests, security scans), and use automation to reduce manual risk. With ML, versioning applies both to code and to model artifacts/metadata (model registry). An organization may include a compliance or governance officer who approves models for production. SageMaker Model Registry supports staged approvals (Pending → Approved/Rejected), enabling explicit sign-off for regulated environments.
An MLOps Engineer slide describing the Compliance Officer role in highly regulated environments. It lists three responsibilities: ensuring ML pipeline alignment with policies, approving or rejecting models for deployment, and monitoring ethical and legal compliance.
SageMaker features commonly used by MLOps engineers:
  • Model Registry: version models and manage approval workflows.
  • SageMaker Pipelines: orchestrate training, validation, registration, and deployment steps.
  • Endpoint deployment: host models for real-time inference (or integrate with other hosting).
  • Model Monitor: continuously detect data or prediction drift and anomalies.
  • SageMaker Clarify: run bias detection and explainability analyses during training and inference.
A slide titled "MLOps Engineer" showing five numbered SageMaker components—Model Registry, Pipelines, Endpoint Deployment, Model Monitor, and Clarify—each with a short description of its role.

​
Governance, Lineage, and Explainability

Governance must be enforced across the entire ML lifecycle—from data ingestion to deployment. Lineage is essential: for any prediction you should be able to trace the dataset, model version, training code, and algorithm that produced it. Capturing lineage enables reproducibility, auditing, and regulatory compliance. Explainability and fairness are critical in regulated or high-stakes domains (finance, healthcare, hiring). Use tools like SageMaker Clarify to run bias detection and produce explainability reports. Model Monitor and telemetry help detect drift in input distributions or prediction quality; pipelines can automatically kick off retraining and redeployment when thresholds are breached.
Enforce governance and automated checks (data validation, fairness tests, security scans, lineage capture) inside CI/CD pipelines. Manual changes are a frequent source of risk—automation reduces errors and improves auditability.
Automation is central: build checks into pipelines so security, data-quality, and fairness tests run whenever code or data changes. When checks pass and approvals are granted, deployment steps proceed as defined—minimizing manual intervention and improving traceability.
By mapping SageMaker capabilities to the roles that use them, teams can design secure, repeatable ML workflows that support rapid experimentation and robust production operations.

Watch Video