1.2.3. Overview of AWS Managed AI/ML Services (Comprehend, Rekognition, etc.)

First Principle: AWS Managed AI Services provide pre-trained intelligence for common tasks, allowing businesses to integrate AI capabilities quickly without needing in-house ML expertise.

For many standard AI tasks, you don't need to build a custom model. You can use an AWS AI Service via a simple API call.

Key Services and Their Use Cases:

• For Vision:
  • Amazon Rekognition: Analyze images and videos to detect objects, faces, text, and inappropriate content.
  • Amazon Textract: Extract text and structured data (from forms and tables) from documents.
• For Language and Speech:
  • Amazon Comprehend: Understand text to find sentiment, entities, and key phrases (NLP).
  • Amazon Translate: Translate text between languages.
  • Amazon Transcribe: Convert speech into text (ASR).
  • Amazon Polly: Convert text into lifelike speech (TTS).
• For Conversational AI:
  • Amazon Lex: Build chatbots and voice assistants.
• For Search and Enterprise Knowledge:
  • Amazon Bedrock Knowledge Bases: A managed retrieval augmented generation (RAG) capability that connects a foundation model to your company's documents and data sources. It retrieves the passages most relevant to a natural language question by meaning, not just matching keywords, and can generate an answer with citations back to the source documents (see Phase 3). (Amazon Kendra, the older enterprise search service, has been closed to new customers since July 30, 2026; AWS recommends Bedrock Knowledge Bases.)
• For Business Applications:
  • Amazon Personalize: Build recommendation engines.
  • Amazon SageMaker Canvas: Build custom prediction models from your own historical data without writing code, such as time-series forecasts or a fraud/not-fraud classifier. (Amazon Forecast and Amazon Fraud Detector, the older single-purpose services, are closed to new customers.)

Agentic AI as a Real-World Application

Beyond single-task services, AI is increasingly applied as agents that complete multi-step work. Examples AWS cites include resolving customer service inquiries and escalating the hard ones to people, speeding up IT incident response, modernizing legacy application code, and anticipating supply-chain delays so deliveries can be rerouted. AWS also offers finished agentic services, such as Amazon Quick for business users and AWS Transform for migration and modernization (see 2.3.3).

Scenario: A company wants to add three features to its app: 1) Automatically moderate user-uploaded profile pictures for inappropriate content. 2) Translate user comments into English. 3) Provide an audio read-out of articles.

Reflection Question: Which three AWS AI Services would you map to these three requirements, and why is using these managed services more efficient than building custom models for these tasks?

💡 Tip: When you see a common business problem like "understanding text" or "analyzing images," your first thought should be, "Is there a managed AWS AI Service for this?" before considering a custom build with SageMaker.