AWS Managed AI Services - Quick Revision Summary
One-page AIF-C01 revision of AWS managed AI services: Comprehend, Translate, Transcribe, Polly, Rekognition, Lex, Personalize, Textract, Kendra and more.
DashboardAWS Managed AI Services - Quick Revision Summary
Amazon Comprehend
- Key Points: Fully managed NLP service that extracts entities, analyzes sentiment, and identifies key phrases from text. Supports custom classification and medical text analysis (Comprehend Medical).
- Main Use Case: Customer interaction analysis, content organization, and extracting insights from unstructured text documents.
Amazon Translate
- Key Points: Deep learning-based language translation service with automatic language detection. Supports custom terminology for brand names and proper nouns.
- Main Use Case: Website/app localization, multilingual content translation, and enabling international user experiences.
Amazon Transcribe
- Key Points: Converts speech to text using ASR (Automatic Speech Recognition). Supports real-time streaming, speaker identification, PII redaction, and custom vocabularies/language models.
- Main Use Case: Call center transcription, subtitling, medical dictation (Transcribe Medical), and converting audio/video to searchable text.
Amazon Polly
- Key Points: Text-to-speech service with multiple voices and languages. Supports SSML for pronunciation control and lexicons for custom word pronunciations.
- Main Use Case: Voice-enabled applications, accessibility features, automated announcements, and generating lifelike speech from text.
Amazon Rekognition
- Key Points: Image and video analysis service that detects objects, people, faces, text, and scenes. Supports custom labels training and content moderation.
- Main Use Case: Facial recognition, content moderation, image labeling, celebrity detection, and video analysis for security and media applications.
Amazon Lex
- Key Points: Natural language chatbot engine that builds conversational interfaces using intents, utterances, and slots. Integrates with Lambda functions to fulfill user requests.
- Main Use Case: Building voice and text chatbots for customer service, ordering systems, and conversational applications.
Amazon Personalize
- Key Points: Fully managed recommendation engine (same as Amazon.com) that provides real-time personalized recommendations via API.
- Main Use Case: Personalized product recommendations, content ranking, and customized marketing for retail and media applications.
Amazon Textract
- Key Points: Extracts text, handwriting, and structured data (forms, tables) from scanned documents using OCR and ML. Supports synchronous (real-time) and asynchronous processing.
- Main Use Case: Document digitization, form processing, invoice extraction, and extracting data from receipts, IDs, and financial documents.
Amazon Kendra
- Key Points: ML-powered intelligent enterprise search that extracts answers from documents using natural language queries. Learns from user interactions to improve results.
- Main Use Case: Enterprise search and FAQ chatbots that provide accurate answers from document collections (S3, Confluence, SharePoint, etc.).
Amazon Mechanical Turk
- Key Points: Managed human task outsourcing marketplace where requesters post HITs (Human Intelligence Tasks) and workers complete them for payment.
- Main Use Case: Data labeling, image classification, and tasks requiring human intelligence that are difficult for ML models.
Amazon Augmented AI (A2I)
- Key Points: Human-in-the-loop service for reviewing low-confidence ML predictions. Integrates with Textract, Rekognition, and SageMaker.
- Main Use Case: Quality assurance for ML predictions by routing uncertain results to human reviewers for validation.
AI Hardware (Trainium & Inferentia)
- Key Points:
- Trainium: ML chip for training large models (100B+ parameters) with 50% cost reduction on Trn1 instances.
- Inferentia: ML chip for inference with 4x throughput and 70% cost reduction on Inf1/Inf2 instances.
- Main Use Case: Cost-effective training and inference for deep learning models at scale.