AWS offers a set of managed AI services, each pre-trained for one specific job, so a business can add speech, language, vision or recommendation features through an API call, without any machine learning expertise. When none of them fits, Amazon SageMaker AI is the platform for building, training and deploying your own models. The exam regularly gives a one-line need and asks which service meets it, so learn each service by what goes in and what comes out.
Speech and Language Services
- Amazon Transcribe converts speech to text: audio in, transcript out. It's used for call recordings, meeting notes and subtitles.
- Amazon Polly converts text to speech: text in, natural-sounding audio out. It's used for reading articles aloud, voice prompts and accessibility.
- Amazon Translate translates text from one language to another: text in, text in another language out.
- Amazon Comprehend finds meaning in text, such as sentiment, key phrases, entities like people and brands, the language used, and personal information. Text in, insights out.
- Amazon Lex builds conversational interfaces, meaning chatbots and voice bots that recognize what a user wants (the intent) and collect the details needed to act on it.
Vision, Document and Recommendation Services
- Amazon Rekognition analyzes images and video: objects, scenes, faces, text in images, and unsafe content.
- Amazon Textract extracts printed and handwritten text, forms and tables from scanned documents. It goes beyond basic optical character recognition (OCR) because it understands the document's structure.
- Amazon Personalize generates personalized recommendations from user behavior data, without building a recommender yourself.
Amazon SageMaker AI
SageMaker AI is the fully managed platform for building, training and deploying your own machine learning models. Choose it when the problem is specific to your business and your data, such as predicting customer churn from your own records, and no pre-trained service does the job. It gives you control over the algorithm, the training data and how the model is deployed, at the cost of needing more ML skill than the pre-trained services. A general rule: start with a pre-trained service if one fits, because it's the fastest and needs no ML expertise, and move to SageMaker AI only when it doesn't.
Chaining Services Together
Real solutions often combine services, so think about the direction of the data:
| Need | Chain |
|---|---|
| Spoken English call to Spanish text | Transcribe, then Translate |
| English article to Spanish audio | Translate, then Polly |
| Call recordings to customer sentiment | Transcribe, then Comprehend |
| Scanned forms to searchable data | Textract, then your database |
On the Exam
Most mistakes come from mixing up opposites and neighbors. Transcribe (speech to text) and Polly (text to speech) are opposites. Comprehend reads meaning from text but doesn't extract text from images; that's Textract. Rekognition analyzes pictures and video, not documents' structure. Lex is the chatbot builder, while Polly only gives it a voice. If the need is custom and based on your own data, the answer is usually SageMaker AI.