Fundamentals of AI and ML · Topic 5 of 8
AWS Managed AI Services
Match business needs to AWS managed AI services such as Rekognition, Comprehend, Textract, Transcribe, Polly, Translate, Lex, Kendra and Personalize.
- 4 min
- Rekognition analyses images and video, Textract extracts text and data from documents, and Comprehend analyses text meaning.
- Transcribe is speech to text, Polly is text to speech, and Translate is language translation.
- Lex builds chatbots and voice bots, and Kendra provides intelligent search over company content.
What it is
AWS offers ready to use AI services that you call through an API. AWS trained and runs the models, so you do not need ML skills, training data or servers to use them.
The exam often describes a business need and asks which service fits. Learn each service by the input it takes and the output it gives.
How it works
The service map
| Service | Input | What it does |
|---|---|---|
| Amazon Rekognition | Images and video | Detects objects, scenes, text and faces, and can flag unsafe content |
| Amazon Textract | Scanned documents and images | Extracts printed text, handwriting, forms and tables |
| Amazon Comprehend | Text | Finds sentiment, key phrases, entities, language and topics (NLP) |
| Amazon Transcribe | Audio | Converts speech to text |
| Amazon Polly | Text | Converts text to lifelike speech |
| Amazon Translate | Text | Translates text between languages |
| Amazon Lex | Voice or text | Builds conversational chatbots and voice bots |
| Amazon Kendra | Documents and questions | Intelligent enterprise search with natural language questions |
| Amazon Personalize | User activity data | Gives personalised recommendations |
Choosing between look alike services
Several services touch the same material, so the exam tests the boundary.
- Textract or Rekognition for text: Textract is built for documents, forms and tables. Rekognition can read text in a scene, such as a street sign in a photo.
- Comprehend or Textract: Textract pulls text out of a scanned page. Comprehend then analyses the meaning of text that already exists.
- Transcribe or Polly: They are opposites. Transcribe turns audio into text, and Polly turns text into audio.
- Translate or Comprehend: Translate converts languages. Comprehend can detect which language a text is written in, but it does not translate.
- Lex or Kendra: Lex holds a conversation to complete tasks, such as booking. Kendra searches company documents to answer questions.
Typical combinations
A call centre can chain services: Transcribe converts calls to text, Comprehend finds the sentiment, and Translate makes the text readable for another team. A document workflow can use Textract to read invoices and Comprehend to classify them.
Human review
Amazon Augmented AI (Amazon A2I) adds human review to ML predictions, for example when the model confidence is low. It lets people check results before they are used, which fits sensitive decisions.
Managed services versus building your own
| Option | Skills needed | Effort | Control |
|---|---|---|---|
| AWS AI service | Low, call an API | Lowest | Limited customisation |
| Amazon Bedrock | Low to medium | Low | Choose and tailor foundation models |
| Amazon SageMaker AI | High | Highest | Full control of the model |
For a common task that a managed service already covers, the managed service is normally the best exam answer, since it is the fastest and needs the least ML work. Build your own model only when the task is unique or needs special data.
Related assistants
Amazon Q is a family of GenAI assistants, covered in the GenAI chapter. It is not a replacement for these task specific services, which give structured results such as labels or extracted fields.
What to remember for the exam
- Rekognition analyses images and video, Textract extracts text and data from documents, and Comprehend analyses text meaning.
- Transcribe is speech to text, Polly is text to speech, and Translate is language translation.
- Lex builds chatbots and voice bots, and Kendra provides intelligent search over company content.
- Personalize produces recommendations from user behaviour.
- Amazon A2I adds human review of ML predictions.
- Managed AI services need no ML expertise and no model training from you.
- When a managed service fits the need, prefer it over building a custom model.
- Move to SageMaker AI when the task is unique and you need full control.
Match the verb in the question to the service: detect objects is Rekognition, extract form fields is Textract, find sentiment is Comprehend, convert speech is Transcribe.
Common mistakes
Using Comprehend to read scanned pages. Comprehend needs text as input. Use Textract first to get the text.
Reversing Transcribe and Polly. Transcribe listens and writes, and Polly reads and speaks.
Picking Lex for document search. Lex handles dialogue. Kendra is the search choice for finding answers in files.
Training a custom model when a managed service exists. This adds cost and time without a clear gain.
Expecting Translate to detect sentiment. Translating is not analysing meaning, so use Comprehend for that.
Check yourself
Which service extracts text, forms and tables from scanned documents?
Amazon Textract.
Which service converts text into lifelike speech?
Amazon Polly.
A company wants employees to ask natural language questions over internal documents. Which service fits?
Amazon Kendra, which provides intelligent enterprise search.
Topic 5 of 8 in Fundamentals of AI and ML
Test yourself on Fundamentals of AI and ML
The full AWS AIF-C01 question bank has 849 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by AWS. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.