AWS AI services explained: Bedrock, SageMaker and more
AWS names most of its AI services after what they do. This guide covers 15 services and concepts for the AI Practitioner exam, including Bedrock versus SageMaker and the common service combinations.
The main idea
AWS named almost every AI service after what it does. Textract extracts text, tables, and form data from scanned documents. Rekognition looks at images and video. Comprehend reads text for sentiment, entities, and key phrases. Transcribe turns audio into text. Translate does real-time language translation. Polly is text to speech, the one name that tells you nothing. Lex builds chatbots.
The two big services show up on almost every practice exam. Bedrock is the AI vending machine. You pick a foundation model such as Claude, Llama, Mistral, or Titan, send prompts, and pay per token. There are no servers and no training. SageMaker is the full ML kitchen. You pick instance types, control the training pipeline, and manage the endpoints.
The exam also asks how services work together. A voice chatbot uses Transcribe as the ears, Lex as the brain, and Polly as the mouth. RAG uses knowledge bases to retrieve documents and Bedrock to generate the answer. Bedrock agents combine instructions, action groups, and knowledge bases, with Lambda functions doing the actions.
Key points
- Extracting data from scanned invoices points to Textract. Image or video analysis points to Rekognition.
- Polly speaks and Transcribe listens. Do not swap them.
- A question that says serverless or foundation model points to Bedrock. Custom training or instance types points to SageMaker.
- Guardrails have four filters: content filters, PII filters, denied topics, and word filters.
- Temperature is the creativity dial. Zero gives the same answer every time and one gives a different answer every time.
Chapters
- 0:00The name game
- 2:35The two big ones: Bedrock and SageMaker
- 3:50The combos
- 5:15Concepts that are not services
Save my free seat →