Integration, data & analytics
AI and machine learning services
SAA-C03 doesn't test ML theory, but it does expect you to recognise managed AI services as the low-effort answer to common needs — "use a purpose-built service" instead of building models yourself (task statement 2.2 even cites Comprehend and Polly).
The services and their clues
| Service | What it does | Clue words |
|---|---|---|
| Amazon Rekognition | Image and video analysis: objects, faces, text, unsafe content, celebrity recognition | "detect objects/faces in images", "content moderation" |
| Amazon Textract | Extract text, forms and tables from scanned documents | "extract data from invoices/forms", "OCR" |
| Amazon Comprehend | Natural language processing: sentiment, entities, key phrases, language, PII detection, topic modelling | "analyse customer reviews", "sentiment" |
| Amazon Transcribe | Speech to text | "transcribe call recordings", "subtitles" |
| Amazon Polly | Text to speech | "convert articles to audio" |
| Amazon Translate | Language translation | "translate content into multiple languages" |
| Amazon Lex | Conversational interfaces — chatbots and voice bots (the technology behind Alexa) | "chatbot", "IVR" |
| Amazon Kendra | Intelligent enterprise search with natural-language questions over documents | "employees search internal documents in natural language" |
| Amazon Forecast | Time-series forecasting (AWS has closed it to new customers — know its purpose) | "forecast demand" |
| Amazon Fraud Detector | Detect online fraud (fake accounts, payment fraud) | "detect fraudulent sign-ups or transactions" |
| Amazon SageMaker | Build, train and deploy custom ML models — notebooks, training, hosting, MLOps | "data scientists build and train their own models" |
Beyond the exam guide, AWS also offers Amazon Bedrock for building generative AI applications with foundation models — useful to know for real-world work, though not listed in SAA-C03.
Combining services: typical pipelines
Call-centre analytics
Call recordings → S3 → Transcribe (speech → text) → Comprehend (sentiment, entities) → Athena / QuickSight
Document processing
Scanned forms → S3 → Lambda → Textract (fields & tables) → validation → DynamoDB → Step Functions for human review
Content moderation
User uploads image → S3 event → Lambda → Rekognition (unsafe content) → approve or quarantine
Key idea
On the exam, if a managed AI service does exactly what's asked, it beats training a model in SageMaker — less effort, faster, cheaper to operate.
Exam patterns
- "Automatically extract amounts and dates from scanned invoices" → Textract.
- "Determine whether customer feedback is positive or negative" → Comprehend.
- "Detect inappropriate images uploaded by users" → Rekognition.
- "Make written content available as audio" → Polly.
- "Build a chatbot for order status" → Lex.
- "Custom model trained on the company's own data" → SageMaker.
Written by Bipin Singh
Work with meSenior Full-Stack Engineer · AI & AWS. I design and run production systems on AWS — serverless, data and AI.