AWS Services for AI, ML and GenAI
Every idea in this course has a matching AWS service. Here is the map, organised by category.
By the end of this lesson you will be able to
- match each major AWS AI service to its category: AI, ML, Deep Learning or Generative AI
- pick the right service for a common use case
- know which services to learn first for certification
Every idea in this course has a matching AWS service. Some are ready-made AI services you call with an API (no ML knowledge needed); others are platforms for building and training your own models; and some are infrastructure such as chips, storage and compute.
Service reference
Scroll the table sideways on a phone.
| AWS service | Category | Purpose | Common use cases |
|---|---|---|---|
| Amazon Bedrock | Generative AI | Build GenAI apps with foundation models, no infrastructure to manage | Chatbots, AI assistants, RAG, summarisation, code and image generation |
| Amazon Nova | Generative AI | AWS's own multimodal foundation models | Text generation, image and video understanding, document processing |
| Amazon Q Business | Generative AI | Enterprise AI assistant | Search company documents, answer employee questions |
| Amazon Q Developer | Generative AI | AI assistant for developers | Code generation, debugging, AWS CLI help, documentation |
| Amazon SageMaker AI | ML / DL / GenAI | Complete platform to build, train, deploy and monitor models | Custom ML models, deep learning, LLM fine-tuning |
| Amazon Rekognition | AI (vision) | Analyse images and videos | Face detection, object detection, PPE detection |
| Amazon Textract | AI (vision) | Extract text and forms from scanned documents | Invoice processing, document digitisation |
| Amazon Comprehend | AI (NLP) | Text analysis | Sentiment analysis, entity recognition, language detection |
| Amazon Comprehend Medical | AI (NLP) | Analyse medical text | Extract diagnoses, medications, procedures |
| Amazon Translate | AI (NLP) | Language translation | Website localisation, multilingual apps |
| Amazon Polly | AI (speech) | Text to realistic speech | Voice assistants, e-learning, phone menus (IVR) |
| Amazon Transcribe | AI (speech) | Speech to text | Meeting transcription, subtitles |
| Amazon Lex | AI (conversation) | Build chatbots and voice bots | Customer support and banking bots |
| Amazon Kendra | AI (search) | Intelligent enterprise search | Internal document search |
| Amazon Personalize | Machine Learning | Personalised recommendations | Netflix-style recommendations, e-commerce |
| Amazon Fraud Detector | Machine Learning | Detect fraudulent activity | Banking, insurance, online payments |
| Amazon Augmented AI (A2I) | AI / ML | Human review of AI predictions | Validate OCR, image recognition, document review |
| AWS Trainium | Deep Learning | Chip specialised for training deep learning models | Large neural network training |
| AWS Inferentia | Deep Learning | Chip specialised for fast, low-cost inference | Production model serving |
| Amazon EC2 GPU instances | DL infrastructure | GPU compute for AI workloads | Neural network training and inference |
| Amazon S3 | ML support | Store datasets and trained models | Training data, model artifacts |
| AWS Glue | ML support | Prepare and transform data (ETL) | Cleaning data before training |
| Amazon EMR | ML support | Process massive datasets | Big-data preprocessing for ML |
| AWS Lambda | AI integration | Run AI workflows serverlessly | Invoke models, automate AI pipelines |
Retired or retiring services you may still see in older courses
AWS regularly retires services. These appear in many older tutorials and exam guides, so it helps to recognise them, but don't build new projects on them. Check the AWS documentation for the latest status.
| Service | What it did | Status |
|---|---|---|
| Amazon Forecast | Time-series forecasting (sales, inventory) | Closed to new customers AWS points users to SageMaker AI (Canvas) instead |
| Amazon Lookout for Metrics | Anomalies in business metrics | Ended Oct 2025 |
| Amazon Lookout for Vision | Product defect detection from images | Ended Oct 2025 |
| Amazon Lookout for Equipment | Predicting equipment failures from sensors | Ending Oct 2026 |
| AWS DeepLens | Camera for learning computer vision | Ended Jan 2024 |
| AWS DeepComposer | Learn generative AI through music | Ended Sep 2025 |
| AWS DeepRacer | Learn reinforcement learning with a model race car | Console ended Dec 2025 now an open-source solution you deploy in your own account |
Category-wise summary
| Category | AWS services |
|---|---|
| Artificial Intelligence (ready-made APIs) | Rekognition, Textract, Comprehend, Comprehend Medical, Translate, Polly, Transcribe, Lex, Kendra |
| Machine Learning | SageMaker AI, Personalize, Fraud Detector |
| Deep Learning | SageMaker AI, Trainium, Inferentia, EC2 GPU instances |
| Generative AI | Bedrock, Nova, Q Business, Q Developer, SageMaker AI (LLM fine-tuning) |
Bedrock is like ordering from a restaurant menu: pick a ready-trained foundation model and call it. SageMaker AI is like a fully equipped kitchen: you bring the data and build, train and serve your own model. Use Bedrock to use GenAI quickly; use SageMaker AI when you need a custom model or full control.
(a) Turn scanned invoices into data. (b) Build a chatbot answering from your company's HR policies. (c) Add voice narration to an e-learning course. (d) Detect whether workers wear helmets in CCTV footage.
Show answers
(a) Textract. (b) Bedrock (with RAG over your documents) or Q Business. (c) Polly. (d) Rekognition (PPE detection).
For AWS certification preparation
These are the services that appear most often in courses and exams. Master them before exploring the specialised ones:
AI
Rekognition, Comprehend, Lex, Polly, Transcribe, Textract
Machine Learning
SageMaker AI, Personalize, Fraud Detector
Deep Learning
SageMaker AI, Trainium, Inferentia
Generative AI
Bedrock, Nova, Q Business, Q Developer, SageMaker AI
Key takeaways
- Ready-made AI services (Rekognition, Comprehend, Polly…) need no ML expertise: call an API.
- SageMaker AI is the do-it-yourself platform; Bedrock is the fastest route to generative AI.
- Trainium speeds up training, Inferentia speeds up inference: the two phases from lesson 4.
Check your understanding
1. You want to call Claude or Amazon Nova from your app without managing servers. Which service?
Bedrock provides managed access to foundation models through one API.
2. Which service converts a recorded lecture into text?
Transcribe is speech-to-text; Polly is the reverse (text-to-speech).
3. AWS Inferentia is designed to speed up…
The name gives it away: it's a chip for serving predictions cheaply and quickly. Trainium is for training.