Chapter Eight
AI & Machine Learning
Two ways to use machine learning on AWS: build your own models, or call a pre-trained API. SageMaker and Bedrock cover the first; the four task-specific APIs cover the second.
The split in this chapter is the practical one. If a pre-trained API solves your problem — reading a document, transcribing audio, detecting objects — use it; it is an API call, not a project.
Reach for SageMaker only when you genuinely need custom models, and for Bedrock when you need generative AI with your own data. Most teams need the APIs far more often than they think they need the platforms.
Services in This Chapter
Service 53
Amazon SageMaker
The end-to-end platform for building, training, and deploying custom ML models. For teams that need their own models, not just an API call.
Service 54
Amazon Bedrock
Managed access to foundation models from multiple providers through one API, with fine-tuning, retrieval, and agents. The generative-AI front door on AWS.
Service 55
Amazon Rekognition
Pre-trained image and video analysis — object, face, text, and moderation detection — through an API call, no model training required.
Service 56
Amazon Comprehend
Pre-trained natural-language processing — sentiment, entities, key phrases, language, and PII — over your text via API.
Service 57
Amazon Textract
Extracts text, forms, and tables from scanned documents. Beyond OCR — it understands structure. The document-ingestion workhorse.
Service 58
Polly, Transcribe & Translate
The speech-and-language trio — text-to-speech, speech-to-text, and machine translation — three pre-trained APIs covered together.