Intermediate

AWS AI/ML Services (36%)

Domain 3 of the AIF-C01 exam — the highest-weighted domain. Master every AWS AI and ML service, know when to use each one, and understand how they fit together to build AI-powered applications.

AWS AI Services Overview

AWS AI services fall into two categories: AI Services (pre-trained, no ML expertise needed) and ML Services (for building custom models). The exam tests your ability to choose the right category and the right service within that category.

Amazon Bedrock (Generative AI)

Bedrock is the central generative AI service on AWS. We covered it in the previous lesson, but here is what matters for this domain:

• Use when: You need text generation, summarization, chatbots, code generation, image generation, or any generative AI task
• Key feature: Access to multiple foundation models (Claude, Titan, Llama, etc.) through one API
• Knowledge Bases: Managed RAG for Q&A over your documents
• Agents: Multi-step task automation with API calling
• Guardrails: Content filtering, PII redaction, denied topics

Amazon SageMaker (Custom ML)

SageMaker is the comprehensive ML platform for building, training, and deploying custom machine learning models. It is the most feature-rich ML service on AWS.

• Use when: You need a custom ML model, the pre-trained AI services do not meet your requirements, or you have specific data and performance needs
• SageMaker Studio — Web-based IDE for ML development
• SageMaker Canvas — No-code ML for business analysts (drag-and-drop model building)
• SageMaker JumpStart — Pre-trained models and solution templates you can deploy with one click
• SageMaker Autopilot — AutoML that automatically trains and tunes models from your data
• Built-in algorithms — XGBoost, Linear Learner, K-Means, and more
• Model deployment — Real-time endpoints, batch transform, serverless inference

Vision Services

Amazon Rekognition

Pre-trained computer vision service for image and video analysis.

• Use when: You need to analyze images or video without building a custom vision model
• Capabilities: Object/scene detection, facial analysis, face comparison, celebrity recognition, text in images (OCR), content moderation (inappropriate content), custom labels (train with your own images), video analysis (people pathing, activity detection)
• Input: Images (JPEG, PNG) or video (stored in S3 or streamed via Kinesis Video Streams)

Amazon Textract

Extract text, forms, and tables from scanned documents.

• Use when: You need to extract structured data from documents (invoices, receipts, forms, IDs)
• Key difference from Rekognition: Textract understands document structure (tables, forms with key-value pairs). Rekognition just reads text in images.
• Capabilities: Plain text extraction, form data extraction (key-value pairs), table extraction, specialized queries ("What is the invoice total?")

Language Services

Amazon Comprehend

Natural Language Processing (NLP) service for text analysis.

• Use when: You need to analyze text for sentiment, entities, key phrases, language, or topics
• Capabilities: Sentiment analysis (positive/negative/neutral/mixed), entity recognition (people, places, dates, organizations), key phrase extraction, language detection, topic modeling, PII detection and redaction
• Comprehend Medical: Specialized version for healthcare text (extracts medical conditions, medications, dosages, procedures)

Amazon Translate

Neural machine translation between languages.

• Use when: You need to translate text between languages in real-time or batch
• Supports: 75+ languages, custom terminology (enforce specific translations for brand names or technical terms), real-time and batch translation

Amazon Transcribe

Automatic speech-to-text (speech recognition).

• Use when: You need to convert audio/video to text
• Capabilities: Real-time and batch transcription, speaker identification, custom vocabularies, automatic language identification, subtitle generation
• Transcribe Medical: Specialized for medical conversations (doctor-patient dialogues)

Amazon Polly

Text-to-speech service that converts text to lifelike audio.

• Use when: You need to convert text to spoken audio
• Capabilities: Multiple voices and languages, Neural TTS (most natural), SSML support (control pronunciation, pauses, emphasis), real-time streaming

Conversational AI

Amazon Lex

Build conversational interfaces (chatbots and voice bots).

• Use when: You need a chatbot or voice bot with structured conversation flows (intents, slots, fulfillment)
• Key concepts: Intents (what the user wants), utterances (how they say it), slots (parameters to collect), fulfillment (action to take)
• Integration: Works with Lambda for business logic, Connect for call centers, and Bedrock for generative responses
• Same technology as Alexa

Business Intelligence & Personalization

Amazon Personalize

Real-time personalization and recommendation engine.

• Use when: You need product recommendations, personalized content, or personalized search results
• Input: User interaction data (clicks, purchases, views), item metadata, user metadata

Amazon Forecast

Time-series forecasting service.

• Use when: You need to predict future values based on historical time-series data (demand forecasting, inventory planning, revenue forecasting)

Amazon Kendra

Intelligent enterprise search powered by ML.

• Use when: You need an enterprise search engine that understands natural language questions and finds answers across documents, FAQs, and knowledge bases
• Difference from Bedrock Knowledge Bases: Kendra returns search results (ranked documents). Bedrock Knowledge Bases generates answers using an LLM. Kendra can feed into Bedrock for a RAG solution.

Other Important Services

Amazon Q

Generative AI assistant for business and development.

• Amazon Q Business — AI assistant that answers questions from company data (documents, wikis, databases)
• Amazon Q Developer — AI coding assistant (code generation, debugging, transformation)

AWS DeepRacer

Autonomous racing car for learning reinforcement learning. Used for education and competitions.

Amazon CodeWhisperer (now Amazon Q Developer)

AI-powered code generation and suggestion tool integrated into IDEs.

Service Selection Cheat Sheet

Practice Questions

Show Answer

B) Amazon Textract. Textract is designed to extract text, forms, and tables from documents. It understands document structure and can extract key-value pairs (invoice number, date) and tables (line items). Rekognition (A) can read text in images but does not understand document structure. Comprehend (C) analyzes text meaning, not document structure. Translate (D) is for language translation.

Show Answer

C) Amazon Rekognition. Rekognition includes content moderation capabilities that detect inappropriate, suggestive, or violent content in images. Comprehend (A) analyzes text, not images. Bedrock Guardrails (B) filters LLM text outputs. Textract (D) extracts text from documents.

Show Answer

C) Amazon Lex. Lex is the conversational AI service that builds chatbots and voice bots. It handles intent recognition, slot filling (collecting information), and fulfillment (routing). Polly (A) converts text to speech but does not understand conversation. Transcribe (B) converts speech to text but does not manage conversation flow. Comprehend (C) analyzes text meaning but does not drive conversations.

Show Answer

B) Amazon SageMaker Canvas. Canvas is the no-code ML tool designed for business analysts. It provides a visual, drag-and-drop interface for building ML models without writing code. SageMaker Studio (A) requires coding experience. Bedrock (C) is for generative AI, not tabular prediction. Comprehend (D) is for NLP, not churn prediction.

Show Answer

B) Amazon Personalize. Personalize is the real-time personalization and recommendation service. It uses user interaction data (browsing, purchases) to generate product recommendations. Kendra (A) is for enterprise search. Forecast (C) is for time-series prediction. Comprehend (D) is for text analysis.