AWS AI services and SageMaker AI tools
The exam rarely asks you to build a model. It asks you to pick the right managed service, so learn the one-line purpose of each.
Pre-built AI services
| Service | Purpose |
|---|---|
| Amazon Rekognition | Image and video analysis, faces, objects, moderation |
| Amazon Textract | Text, forms and tables from documents |
| Amazon Comprehend | NLP: sentiment, entities, key phrases, PII detection |
| Amazon Transcribe | Speech to text |
| Amazon Polly | Text to speech |
| Amazon Translate | Language translation |
| Amazon Lex | Chatbots and voice bots |
| Amazon Kendra | Intelligent enterprise search |
| Amazon Personalize | Recommendations |
| Amazon Neptune | Graph database, used for graph-based ML |
The pattern: pick a pre-built service when the task is common and you do not want to train a model. Use SageMaker AI when you need custom models.
Amazon SageMaker AI
Amazon SageMaker was renamed Amazon SageMaker AI; the exam may use either name. Key features:
| Feature | Use |
|---|---|
| Studio | Integrated development environment for ML |
| Canvas | No-code model building for analysts |
| Data Wrangler | Prepare and transform data |
| Ground Truth | Data labeling with human reviewers |
| JumpStart | Pre-trained models and solution templates |
| Pipelines | Automate ML workflows (MLOps) |
| Model Registry | Store, version, approve models |
| Model Monitor | Detect data and quality drift in production |
| Clarify | Bias detection and explainability |
| Model Cards | Document a model’s purpose, risks and training |
| Feature Store | Share and reuse features |
Deployment options
SageMaker AI offers real-time endpoints, serverless inference (no infrastructure to manage, intermittent traffic), asynchronous inference (large payloads), and batch transform (offline scoring of whole datasets).
Exam traps
- Textract extracts structure from documents; Rekognition analyzes pictures; Comprehend analyzes text meaning.
- Polly speaks, Transcribe listens, Translate converts languages.
- Model Monitor watches a deployed model for drift; Clarify checks bias and explainability; Model Cards document; Model Registry versions.
- Canvas is no-code, JumpStart is a catalog of pre-trained models, Studio is for developers.
- Lex builds the conversation; it does not generate free-form answers like a foundation model.
Practice questions
Choose an answer, then review the explanation to check your understanding.
1. A company needs to pull text, tables and form fields from scanned invoices. Which service should it use?
Show answer
Answer: B.Amazon Textract extracts text, forms and tables from scanned documents.
2. A team wants to add subtitles and spoken voice-overs in several languages to its videos. Which services combine to do this?
Show answer
Answer: B.Transcribe converts speech to text, Translate translates it, and Polly turns text into speech.
3. Which SageMaker AI feature stores, versions and approves trained models before deployment?
Show answer
Answer: A.Model Registry catalogs and versions models and tracks approval status.
4. A business analyst with no coding skills wants to build a prediction model from a spreadsheet. Which tool fits?
Show answer
Answer: A.SageMaker Canvas is a no-code visual interface for building ML models.