AWS Machine Learning Services: What Each One Does

AWS offers a family of ready-made machine learning services that handle a single, well-defined task each: recognize faces, transcribe speech, translate text, recommend products, and so on. You call an API and get a result — no model training, no GPUs, no data science required. SageMaker sits alongside them for the cases where you do want to build your own model.

This is a tour of what each service is, what it is for, and when to reach for it.

The mental map

┌─────────────────────────────────────────────────────────┐
│  VISION (images, documents)                             │
│  • Rekognition  (objects, faces, scenes in images/video)│
│  • Textract     (text & data from scanned documents)    │
├─────────────────────────────────────────────────────────┤
│  SPEECH (audio in, audio out)                           │
│  • Transcribe   (speech → text)                         │
│  • Polly        (text → speech)                         │
├─────────────────────────────────────────────────────────┤
│  LANGUAGE (text understanding)                          │
│  • Translate    (translation across languages)          │
│  • Comprehend   (NLP — sentiment, entities, topics)     │
├─────────────────────────────────────────────────────────┤
│  CONVERSATIONAL                                         │
│  • Lex          (chatbots, voice bots)                  │
│  • Connect      (cloud contact center, pairs with Lex)  │
├─────────────────────────────────────────────────────────┤
│  SEARCH & RECOMMENDATIONS                               │
│  • Kendra       (ML-powered document search)            │
│  • Personalize  (real-time recommendations)             │
├─────────────────────────────────────────────────────────┤
│  BUILD YOUR OWN                                         │
│  • SageMaker    (full ML platform for data scientists)  │
└─────────────────────────────────────────────────────────┘

Amazon Rekognition

Rekognition finds objects, people, text, and scenes in images and videos using machine learning. It also does facial analysis and facial search — useful for user verification and people counting — and can match against a database of “familiar faces” or against known celebrities.

Typical use cases:

• Labeling — auto-tag images by what is in them.
• Content moderation — flag unsafe or inappropriate content.
• Text detection — pull text out of an image.
• Face detection and analysis — gender, age range, emotions.
• Face search and verification.
• Celebrity recognition.
• Pathing — track how a person or object moves through a video, for example in sports analysis.

Amazon Transcribe

Transcribe converts speech to text. Under the hood it uses a deep learning process called automatic speech recognition (ASR) to do it quickly and accurately.

A few features worth knowing:

• Redaction automatically removes Personally Identifiable Information (PII) from the transcript.
• Automatic Language Identification handles multi-lingual audio.

Common uses:

• Transcribing customer service calls.
• Generating closed captions and subtitles.
• Producing searchable metadata for a media archive.

Amazon Polly

Polly is the mirror image of Transcribe: it turns text into lifelike speech using deep learning. It is what you reach for when you want an application that talks — voice assistants, IVR, accessibility features, narrated content.

Amazon Translate

Translate provides natural and accurate language translation. It lets you localize content — websites, applications, documents — for international users, and translate large volumes of text without writing your own translation pipeline.

Amazon Lex & Connect

These two often appear together because they build the same thing — a voice-driven application — at different layers.

Amazon Lex is the same technology that powers Alexa. It combines:

• Automatic Speech Recognition (ASR) to convert speech to text.
• Natural Language Understanding to recognize the intent behind that text.

It is what you use to build chatbots and call-center bots.

Amazon Connect is a cloud-based virtual contact center. It receives calls, lets you design contact flows, and integrates with CRMs and other AWS services. AWS markets it as roughly 80% cheaper than traditional contact-center solutions, with no upfront payments.

Together they look like this:

Phone Call                 call         stream        invoke       schedule
"Schedule an    ──►  Connect  ──►  Lex (intent  ──►  Lambda  ──►  CRM
 Appointment"                       recognized)

Connect receives the call, Lex figures out what the caller wants, Lambda executes the action, and the CRM is updated.

Amazon Comprehend

Comprehend is the natural language processing (NLP) service — fully managed and serverless. It uses machine learning to find insights and relationships in text:

• The language of the text.
• Key phrases, places, people, brands, and events.
• Sentiment — how positive or negative the text is.
• Tokenization and parts of speech.
• Topic modeling — automatically organize a collection of text files by topic.

Sample uses:

• Analyzing customer emails to see what leads to positive or negative experiences.
• Grouping articles by topics that Comprehend discovers on its own.

Amazon SageMaker

The previous services are all pre-trained — you call them and get an answer. SageMaker is the opposite: it is a fully managed platform for developers and data scientists to build, train, and deploy their own ML models.

Without SageMaker, doing the whole ML lifecycle in one place — provisioning servers, labeling data, training, tuning, deploying — is awkward. SageMaker rolls it into a single service. The simplified flow:

Historical data ──label──► labeled data ──build──► ML model ──train & tune
                                                       │
                                                       ▼
                                                  New data ──apply──► Prediction

If you have a custom problem and want to train a model on your own data, SageMaker is the place. If a problem fits Rekognition, Transcribe, Comprehend, or one of the other purpose-built services, use those instead — they are simpler and you don’t need to train anything.

Amazon Kendra

Kendra is a fully managed document search service powered by machine learning. It can extract answers from inside documents — text, PDFs, HTML, PowerPoint, Word, FAQs — and answer questions in natural language.

It connects to a wide range of data sources (S3, RDS, Google Drive, SharePoint, OneDrive, Salesforce, ServiceNow, and more), indexes them, and exposes a unified search interface.

Two features worth calling out:

• Incremental learning — it learns from user interactions and feedback to promote preferred results.
• Manual tuning — you can boost importance, freshness, or specific results yourself.

Picture an employee typing “Where is the IT support desk?” into a search bar and Kendra answering “1st floor” by pulling it from an internal wiki.

Amazon Personalize

Personalize is a managed service for real-time personalized recommendations — the same technology that powers Amazon.com. You feed it user and item data; it returns recommendations through a customized API.

It plugs into existing websites, apps, SMS, and email systems. The selling point: you can ship recommendations in days rather than spending months building, training, and deploying your own model.

Amazon S3 ──read data──►
                         Amazon Personalize ──Customized API──► Websites, apps,
Personalize API ──stream─►                                       SMS, emails

Typical use cases are retail stores, media, and entertainment — anywhere “users who liked X also liked Y” creates value.

Amazon Textract

Textract automatically extracts text, handwriting, and data from scanned documents using AI and ML. It handles PDFs and images, and it understands the structure of forms and tables — so it doesn’t just give you a wall of text, it gives you fields and values.

A scanned driver’s license, for example, comes back as JSON:

{
  "Document ID": "123456789-005",
  "Name": "...",
  "SEX": "F",
  "DOB": "23.05.1997"
}

Common domains:

• Financial services — invoices, financial reports.
• Healthcare — medical records, insurance claims.
• Public sector — tax forms, ID documents, passports.

Decision shortcuts

Summary

• AWS’s ML services split cleanly into two camps: purpose-built APIs that solve one well-defined task each, and SageMaker for building your own models.
• For vision, reach for Rekognition and Textract. For speech, Transcribe and Polly. For language, Translate and Comprehend. For conversation, Lex with Connect on top.
• Kendra is search, Personalize is recommendations, and SageMaker is the escape hatch when none of the pre-trained services fit.
• Pick by the task you want to solve — the service names are descriptive enough that the right one is usually obvious once you know they exist.