Skip to main content
Welcome. This lesson summarizes the core AWS AI/ML services you should know for architecture decisions and certification exams. Focus on each service’s primary purpose, key capabilities, and common integrations—those are the exam-relevant points.
This lesson highlights many high-level AWS AI/ML services and common integration patterns. You don’t need to memorize every single subfeature, but you should know the main purpose and key capabilities for each service.
AWS offers both pre-trained AI services (APIs you can call immediately) and platforms/infrastructure for building, tuning, and operating custom ML models. Examples covered below include: SageMaker, Polly, Lex, Rekognition, Bedrock, Transcribe, Textract, Comprehend, Translate, Fraud Detector, Glue, DataBrew, EMR, A2I, and QuickSight.
A dark-blue slide titled "AWS Is a Leader in AI Services" listing three services—AWS SageMaker, AWS Polly, and AWS Lex—with brief descriptions and green icons. The items are presented as horizontal cards with colored side accents.
Common reasons teams choose AWS for AI/ML: global scalability, pay-as-you-go pricing, built-in high availability, and a security posture integrated with the broader AWS platform. Service categories (high level)
  • Pre-trained AI services: ready-to-use APIs for vision, speech, language, and document extraction (e.g., Rekognition, Transcribe, Translate, Comprehend, Textract).
  • ML platforms: model development, training, and deployment tooling (e.g., SageMaker).
  • Infrastructure for custom AI: compute, storage, and managed frameworks for training and serving (EC2/GPU instances, EMR, Bedrock for managed foundation models).
A slide titled "AWS AI/ML – Service Categories" showing three columns: 01 Pre-trained AI services, 02 ML platforms, and 03 Infrastructure for custom AI. Each column includes brief bullet points like ready-to-use APIs, tools for model development, and training infrastructure/GPU instances.
Quick reference table — service, primary purpose, and common integrations:

​
Amazon Bedrock — Foundation models (managed)

Amazon Bedrock provides managed access to diverse foundation models from AWS, third-party, and open-source providers. It’s designed to simplify building generative AI applications without the operational burden of provisioning and maintaining model infrastructure. Key capabilities:
  • Access to multiple foundation models and vendors through a single API.
  • Customization: fine-tuning or prompt-tuning with your data.
  • Governance & safety features (moderation, policies) and orchestration via Bedrock Agents for multi-step workflows.
Integrations: S3 for data, SageMaker for custom training/experiments, Lambda and EC2 for application logic and hosting.
The image is a slide titled "What Is AWS Bedrock?" It states that Amazon Bedrock is a fully managed service for foundation models that simplifies generative AI application development.
A slide titled "AWS Bedrock – Capabilities" on a dark background listing three features: access to diverse foundation models, easy customization with your data, and seamless integration with AWS services. Colored vertical bars accent each feature.
A presentation slide titled "AWS Bedrock – Guardrails" listing three features: "Ensuring safe and responsible AI," "Built-in content moderation," and "Compliance and data privacy features."

​
Amazon SageMaker — End-to-end ML platform

SageMaker is the primary AWS platform for custom ML development, training, tuning, and deployment. It targets data scientists and ML engineers while providing features that reduce operational complexity for teams. What SageMaker provides:
  • Ground Truth for data labeling and human-in-the-loop workflows.
  • SageMaker Studio (IDE) and managed Jupyter notebooks.
  • Data Wrangler for data preparation and visualization.
  • Scalable training jobs (distributed training, spot instances).
  • Hyperparameter tuning and model evaluation tools.
  • Managed real-time endpoints and batch transform for offline inference.
  • Model monitoring, versioning, and deployment strategies (A/B, blue/green).
Typical SageMaker workflow:
  1. Data ingestion (S3, EFS).
  2. Preparation and exploration (Notebooks, Data Wrangler).
  3. Training (distributed jobs).
  4. Evaluation & tuning.
  5. Deployment (Endpoints).
SageMaker also integrates with ETL services (Glue), monitoring (CloudWatch), and human-review (A2I).
A presentation slide titled "Key Features" showing five numbered dark tiles with icons listing: Data labeling tools, Notebooks for model development, Automatic model tuning, Scalable deployment, and Easy-to-use for those with minimal AI expertise.
A horizontal diagram titled "SageMaker Workflow" showing five numbered steps: 01 Data Ingestion, 02 Data Preparation and Exploration, 03 Model Training, 04 Model Evaluation and Tuning, and 05 Model Deployment. Each step lists related tools/services beneath (e.g., Amazon S3, SageMaker Notebook/Data Wrangler, Training Jobs, Automatic Model Tuning, SageMaker Endpoints).
A diagram of an AWS SageMaker workflow showing datasets in an S3 bucket passed through AWS Glue for ETL and then into SageMaker for exploratory data analysis, data cleaning, model building, and deployment.
Notes:
  • Supports bring-your-own-algorithm and built-in algorithms.
  • Works with many instance types (including GPU) and cost-saving spot instances.
  • SageMaker Canvas enables no-code model building for business users.

​
Amazon Rekognition — Computer vision

Rekognition provides image and video analysis: object/scene detection, facial analysis & recognition, text detection, celebrity recognition, custom labels, emotion inference, content moderation, and near-real-time stream analysis. Use cases: automated tagging, visual search, media analysis, security monitoring. Integrations: S3, Lambda, DynamoDB, Kinesis Video Streams, A2I for human review.
A dark presentation slide titled "What Is Amazon Rekognition?" It states that Amazon Rekognition analyzes images and videos for object detection, facial analysis, and text recognition.
A presentation slide titled "Features" showing five dark cards numbered 06–10. Each card lists a capability: Celebrity recognition, Custom labels, Integration with other AWS services, Emotion detection, and Real-time analysis.

​
Amazon Lex — Conversational interfaces

Lex combines automatic speech recognition (ASR) and natural language understanding (NLU) to build chatbots and voice assistants. It interprets intent and slots from text or speech, and triggers business logic via Lambda. Common integrations: Amazon Cognito (auth), Lambda (business logic), Polly (TTS), and databases like DynamoDB or DocumentDB for session/state. Use cases: customer support bots, booking systems, voice-enabled applications.
A simple AWS architecture diagram titled "Lex" showing a user authenticating with Amazon Cognito and interacting with an Amazon Lex chatbot. Lex calls an AWS Lambda function which reads/writes data to Amazon DocumentDB.

​
Amazon Polly — Text-to-speech (TTS)

Polly generates lifelike speech from text and supports SSML to control pronunciation, emphasis, and prosody. It can stream audio in real time or produce files saved to S3. Common patterns: pair with Lex or Transcribe in voice flows, use Translate + Polly for multilingual voice responses, or orchestrate via Lambda/S3.
A minimalist illustration titled "Polly" showing a person icon on the left and a robot icon on the right exchanging speech bubbles. The conversation bubble in the center shows the user asking about the weather and the robot replying with temperature and forecast details.
A standard voice workflow:
  • User speech -> Transcribe (speech-to-text) -> business logic (Lambda / IoT) -> Polly (text-to-speech) -> spoken response.
Diagram of an AWS Polly workflow. A text file uploaded to an input S3 bucket triggers a Lambda function that sends the text to Amazon Polly for text-to-speech and saves the resulting audio to an output S3 bucket.
A dark-themed flow diagram titled "Polly" showing a voice command ("Set the temperature to 22 C") converted by Amazon Transcribe, processed by AWS Lambda and AWS IoT Core, and used to control a smart thermostat with a spoken response generated by Amazon Polly. Icons for a user, Alexa/assistant, and AWS services illustrate the end-to-end voice-to-device workflow.

​
Amazon Comprehend — Natural language processing

Comprehend performs language detection, sentiment analysis, entity recognition, key-phrase extraction, syntax analysis, topic modeling, and PII detection. It’s suitable for analyzing customer feedback, social media streams, emails, and other text corpora. Typical pipeline: text stored in S3 -> Lambda triggers Comprehend -> results stored for analytics (Athena, QuickSight, or databases).

​
Amazon Fraud Detector — Fraud detection workflows

Fraud Detector provides a managed workflow for detecting fraudulent activity using ML models, including:
  • Defining use cases and events,
  • Ingesting historical data,
  • Training and evaluating models,
  • Deploying detectors for real-time or batch scoring,
  • Applying decision logic (approve/review/decline).
It integrates with S3, Lambda, and A2I (for human review) and supports feedback loops to retrain models.
A slide titled "Fraud Detector" showing nine numbered dark rounded boxes that outline the steps to build a fraud detection system. The steps include defining the business use case, inputting historical data, selecting and training models, generating scores and evaluating performance, adjusting and retraining, establishing decision logic, deploying the detector, running real-time or batch evaluations, and assigning outcomes.
A flow diagram titled "How Fraud Detector Works" showing a client application sending requests to a fraud detector model that returns risk scores. Suspicious cases are routed to human reviewers and stored in Amazon S3, with results feeding back into the fraud detector.

​
Amazon Transcribe — Speech-to-text

Transcribe converts audio to text for both real-time streaming and batch jobs. Features include custom vocabularies, speaker diarization, and language support. Commonly used in media processing, meeting transcription, and downstream NLP analysis. Typical flow: audio/video in S3 -> Lambda triggers Transcribe -> transcribed text -> optional Translate/Comprehend -> storage or analytics.
A slide titled "Features" with three dark cards numbered 01–03 showing icons and the labels "Automatic speech recognition," "Real-time transcription," and "Custom vocabulary." The slide also shows a small "© Copyright KodeKloud" in the bottom-left.
A slide titled "Transcribe – Speaker Diarization" showing two blue user icons labeled spk_0 and spk_1 with an audio waveform between them and an arrow pointing from the left icon to the right. It illustrates the flow of speaker diarization in a transcription process.
A schematic titled "Transcribe" showing an AWS transcription workflow with icons for Amazon S3, AWS Lambda, and Amazon Transcribe connected by arrows. The transcribed output is then routed to Amazon Translate, Amazon Comprehend, and Amazon DynamoDB on the right.

​
Amazon Translate — Neural machine translation

Translate provides real-time and batch neural machine translation. It supports custom terminology for domain-specific vocabulary and integrates well with Transcribe (speech→text→translate) and Polly (speak translations). Common flow: audio -> Transcribe -> Translate -> Polly (optional TTS output).
A slide titled "Features" showing five numbered feature cards: 01 Neural Machine Translation, 02 Wide range of supported languages, 03 Real-time translation, 04 Seamless integration, and 05 Custom terminology.
A simple AWS workflow diagram titled "Translate" showing data flowing from Amazon S3 to Amazon Translate and then to Amazon Comprehend, with icons for each service.

​
Amazon Textract — Document OCR & structured data extraction

Textract extracts text, tables, and form fields (OCR) from scanned documents, forms, and handwriting. Outputs can be structured JSON, stored to S3, or pushed to databases for downstream processing or analysis. Standard workflow: document in S3 -> Lambda or direct API invoke -> Textract -> extracted structured data -> storage/processing.
A flow diagram titled "Textract" showing data moving from Amazon S3 to AWS Lambda, then to Amazon Textract, and finally into Amazon DynamoDB. Each service is represented by its icon in a left-to-right pipeline.

​
Data & ETL: AWS Glue, DataBrew, EMR

  • AWS Glue: serverless ETL and data cataloging. Crawls sources to populate the Data Catalog, runs PySpark/Scala ETL jobs, and writes processed data to S3, Redshift, etc. Author jobs in Glue Studio or with custom code.
  • Glue DataBrew: visual data cleaning and transformation tool (no code). Create recipes of transformations and run jobs on datasets.
  • Amazon EMR: managed big data platform (Hadoop, Spark, Hive, etc.) for large-scale data processing. Supports cluster and serverless operation modes.
A diagram titled "Glue DataBrew" showing an AWS data pipeline: AWS Glue DataBrew and AWS Glue process data into a central store, which is queried by Athena and visualized in QuickSight. The outputs are consumed by a Data Analyst and a Business Analyst.
A diagram titled "Elastic MapReduce (EMR)" showing a processing pipeline with steps: "Submit input dataset," "Process with Pig program," "Process with Hive program," and "Write output dataset," plus status boxes below indicating states like "Begin processing," "Completed," "Failed," and "Cancelled." The slide is dark-themed and credited to KodeKloud.

​
Amazon Augmented AI (A2I) — Human-in-the-loop

A2I provides human review workflows for cases where model confidence is low or compliance requires human validation. It integrates with SageMaker, Textract, Rekognition, and other services. Human feedback can be fed back into training datasets to improve models. Common use cases: sensitive PII handling, manual verification, quality control.
A presentation slide titled "Features" showing five numbered cards with icons. The cards list: 01 Easy integration with ML services; 02 Built-in Human Review workflows; 03 Access to Human Reviewers; 04 Continuous learning and improvement; 05 Seamless scaling.

​
Amazon QuickSight — BI & visualization

QuickSight is a serverless BI tool for dashboards, charts, and visual analytics. It connects to Athena, Redshift, Glue Data Catalog, and other sources and uses SPICE (in-memory engine) for fast queries. QuickSight supports natural-language Q&A (Amazon Q). Use case: visualize model metrics, business KPIs, and analytics results from ML pipelines.
When working with PII, medical data, or other sensitive information, apply encryption at rest/in transit, IAM least privilege, appropriate guardrails (Bedrock), and human-review workflows (A2I). Ensure compliance with regulations (HIPAA, GDPR) where applicable.
Study tips and exam focus
  • Learn each service’s primary purpose and common integrations rather than minutiae.
  • For architecture questions: identify whether a solution uses a pre-trained API (Rekognition, Comprehend, Translate) or requires custom modeling (SageMaker + EC2/EMR).
  • Remember common pipelines: S3 → Lambda → service → S3/DB → analytics (Athena/QuickSight).
  • Know which services support streaming vs batch (e.g., Transcribe supports real-time streaming).
Further reading and references That covers the primary AWS AI/ML and supporting data services for this lesson. Focus on the purpose, typical integrations, and example pipelines for each service to prepare effectively for architecture and exam scenarios. Good luck with your studies.

Watch Video