AWS AI/ML Application Services
AWS offers a suite of pre-trained, fully managed AI services that let you add intelligence to applications without any machine learning expertise. These services are API-driven — you send data in, get predictions back. When your task matches what these services do, they are almost always the fastest and simplest path to production.
Natural Language Processing (NLP)
| Service | What It Does | When to Use |
|---|---|---|
| Amazon Comprehend | Sentiment analysis, entity recognition, key phrases, language detection, topic modeling, PII detection | Text analytics without building custom models. Comprehend Custom Classifier enables text classification with your own labeled data |
| Amazon Transcribe | Speech-to-text (automatic speech recognition) | Convert audio or video to text. Supports multiple languages |
| Amazon Translate | Neural machine translation between languages | Translate text between any supported language pair |
| Amazon Polly | Text-to-speech with lifelike voices | Generate spoken audio from text. Supports SSML for pronunciation control and Neural TTS for natural-sounding speech |
| Amazon Lex | Conversational AI — chatbots and voice bots | Build chatbots and voice interfaces. Uses the same technology as Alexa (Intents + Slots + Lambda fulfillment) |
The NLP Pipeline Pattern
A common architecture chains NLP services together:
Audio (any language) → Transcribe (speech to text) → Translate (to target language) → Comprehend (sentiment, entities, topics)
This pattern handles multilingual audio analysis end-to-end with no custom ML.
Computer Vision
| Service | What It Does | When to Use |
|---|---|---|
| Amazon Rekognition | Image/video analysis: face detection, object detection, celebrity recognition, content moderation, text in images, PPE detection | Image or video analysis without training custom models. Rekognition Custom Labels enables custom image classification |
| Amazon Textract | Extract text, tables, and forms from scanned documents and PDFs | Document processing, invoice extraction, form parsing. Goes beyond OCR — understands document structure (tables, key-value pairs) |
Rekognition Key APIs
| API | Purpose |
|---|---|
| CompareFaces | Compare two faces for similarity |
| IndexFaces | Add faces to a collection for later search |
| SearchFaces | Search a face collection for matches |
| Custom Labels | Train a custom image classifier with your own labeled images |
Forecasting and Recommendations
| Service | What It Does | When to Use |
|---|---|---|
| Amazon Forecast | Managed time-series forecasting | Demand forecasting, resource planning, financial planning. Requires separate CSVs: target time series, related time series, and item metadata |
| Amazon Personalize | Real-time personalized recommendations | Product recommendations, content personalization. Use Event Tracker for real-time user interactions without full retraining |
Amazon Forecast Details
- CNN-QR algorithm supports related time series data and holiday calendars
- Uses weighted Quantile Loss (wQL) — set higher quantiles when underforecasting is more costly than overforecasting
- Requires structured input: target time series (required), related time series (optional), item metadata (optional)
Specialized AI Services
| Service | What It Does | When to Use |
|---|---|---|
| Amazon Fraud Detector | Detect online fraud using your data | Payment fraud, fake accounts, account takeover |
| Amazon Kendra | Intelligent enterprise search powered by ML | Search across documents, FAQs, and knowledge bases with natural language queries (not the same as OpenSearch — Kendra understands questions) |
| Amazon Bedrock | Managed foundation models (Claude, Titan, Llama, etc.) | Generative AI, text generation, summarization, RAG. Serverless access to foundation models |
| Amazon Q | AI assistant for business — answers questions from enterprise data | Enterprise Q&A, business intelligence, code assistance |
| Amazon A2I (Augmented AI) | Human review of ML predictions in production | Human-in-the-loop for low-confidence predictions |
| Amazon Lookout for Equipment | Predictive maintenance using sensor data | Detect equipment anomalies from sensor time-series. Used for inference, not data cleaning |
| Amazon Mechanical Turk | Crowdsourced human intelligence tasks | Data labeling workforce for non-sensitive data. Integrated with SageMaker Ground Truth |
When to Use
Choose managed AI services when:
- The task matches what the service does (text analytics, image analysis, translation, etc.)
- You want the fastest path to production
- Your team does not have ML expertise
- You do not need custom model architectures
Use SageMaker instead when you need custom algorithms, fine-grained hyperparameter control, or when the problem does not fit any pre-built service.
Flashcards
What is the NLP pipeline pattern for multilingual audio analysis?
Click to revealTranscribe (speech→text) → Translate (→target language) → Comprehend (sentiment/entities/topics). Three managed services, no custom ML.
When the requirement is "least effort" or "no ML expertise needed," managed AI services (Comprehend, Rekognition, Textract, Forecast, Personalize) are almost always the right answer. Reserve SageMaker for scenarios that explicitly require custom model training or built-in algorithms.