10. Machine Learning

AWS provides pre-trained AI services for developers and a comprehensive platform (SageMaker) for data scientists.

Quick-Recall Decision Table (exam-critical): SAA-C03 rarely asks you to explain how any of these services work internally — it describes a scenario in plain English and expects you to match it to the service by input/output type alone.

Input	Output / Task	Service
Plain/unstructured text	Sentiment, entities, key phrases	Amazon Comprehend
Scanned documents (forms, tables)	Structured text/data extraction (beyond OCR)	Amazon Textract
Images/video	Object, face, or inappropriate-content detection	Amazon Rekognition
Audio/speech	Text transcript	Amazon Transcribe
Text	Spoken audio	Amazon Polly
Text in one language	Text in another language	Amazon Translate
User utterances (voice/text)	Conversational chatbot / IVR	Amazon Lex
Natural-language question over internal documents	Enterprise search / Q&A	Amazon Kendra
Custom/tabular business data	Trained, deployable custom ML model	Amazon SageMaker AI

The tell: if the scenario just needs a pre-built, no-training-required capability (moderate content, transcribe a call, translate a page), it's one of the single-purpose AI services above. If it says "train a custom model on our own historical/business data," that's always SageMaker — none of the single-purpose services accept custom training data in the same way.

Amazon Comprehend

Service Introduction: A natural language processing (NLP) service that uses ML to find insights and relationships in text.

Common Usage: Analyzing text for sentiment, key phrases, entities, or language without ML expertise.

Project Examples:

• Automating sentiment analysis on thousands of customer support tickets.
• Extracting medical entities from unstructured clinical notes.

Amazon Kendra

Service Introduction: An intelligent search service powered by ML that provides highly accurate search results from unstructured data sources.

Common Usage: Building enterprise search engines that can answer natural-language questions from internal wikis and documents.

Project Examples:

• Creating an internal HR portal where employees can ask "How do I enroll in benefits?"
• Implementing a search engine for a technical documentation site.

Amazon Lex

Service Introduction: A service for building conversational interfaces (chatbots) into any application using voice and text.

Common Usage: Powering the same conversational technology as Alexa to automate customer service interactions.

Project Examples:

• Building an automated chatbot for an airline’s flight-booking website.
• Creating a voice-controlled IVR system for a banking call center.

Amazon Polly

Service Introduction: A service that turns text into lifelike speech, allowing you to create applications that talk.

Common Usage: Adding voice capabilities to apps or creating audio versions of written content to increase accessibility. Key Sub-Features (exam-critical):

• Custom Lexicons — Define custom pronunciations for domain-specific terms (product names, abbreviations, acronyms) so they're spoken correctly. This is the standard answer whenever a question mentions "custom product names/abbreviations."
• StartSpeechSynthesisTask — The asynchronous API operation for converting long-form text (e.g., each section of a manual) to speech with the lowest operational overhead, vs. the synchronous SynthesizeSpeech call.

Exam Example (SAA-C03 pattern): "A developer is building a text-to-speech application and needs to ensure that specific product abbreviations are pronounced correctly by the service. Which feature of Amazon Polly should be utilized?" Answer choices: StartSpeechSynthesisTask / Standard Voices / Custom Lexicons / Speech Synthesis Markup Language (SSML).

• Correct Answer: Custom Lexicons — Let you define specific pronunciations for abbreviations, acronyms, or domain-specific terms so Polly speaks them correctly.
• Why not StartSpeechSynthesisTask? — It's the async API operation for long-form text conversion, not a feature for defining pronunciations.
• Why not Standard Voices? — They provide the speech engine but don't inherently know how to handle unique company-specific abbreviations without a Lexicon.
• Why not SSML? — SSML can adjust pronunciation inline per-request, but Lexicons are the standard answer for applying a custom pronunciation globally and consistently across an entire workload.

Project Examples:

• Converting news articles into downloadable audio podcasts.
• Adding real-time voice notifications to an automated public announcement system.

Amazon Rekognition

Service Introduction: A computer vision service that makes it easy to add image and video analysis to applications.

Common Usage: Detecting objects, faces, text, and inappropriate content in images and video streams.

Project Examples:

• Implementing automated content moderation for user-uploaded profile photos.
• Building a facial-recognition-based secure access system for an office.

Amazon SageMaker AI

Service Introduction: A fully managed service that provides developers and data scientists the ability to build, train, and deploy ML models at scale.

Common Usage: Managing the entire ML lifecycle, including data labeling, notebook hosting, and high-performance model endpoints. SageMaker Notebooks (sub-feature, exam-critical distinction): A managed Jupyter notebook environment where data scientists write Python or R code to explore data, experiment, and transform datasets. Unlike Glue DataBrew, this is a code-first tool — it's the wrong answer whenever a question specifically asks for a code-free, visual data-prep interface for business analysts.

Project Examples:

• Training a custom fraud detection model using historical transaction data.
• Hosting a real-time product recommendation engine for an e-commerce site.

Amazon Textract

Service Introduction: A service that automatically extracts text, handwriting, and data from scanned documents beyond simple OCR.

Common Usage: Digitizing documents like invoices or medical records while maintaining table and form structures.

Project Examples:

• Automating the ingestion of data from standard tax forms into a database.
• Extracting key-value pairs from thousands of handwritten legal contracts.

Amazon Transcribe

Service Introduction: An automatic speech recognition (ASR) service that makes it easy to add speech-to-text capability to applications.

Common Usage: Converting audio files or live speech into text for analysis or captioning.

Project Examples:

• Generating automated subtitles for live-streamed webinars.
• Transcribing customer service calls to perform text-based sentiment analysis.

Amazon Translate

Service Introduction: A neural machine translation service that delivers fast, high-quality, and affordable language translation.

Common Usage: Localizing application content and enabling real-time communication between users speaking different languages.

Project Examples:

• Translating a global knowledge base into 20 different languages.
• Providing real-time chat translation for a global multiplayer game.