1.2.3. Overview of AWS Managed AI/ML Services (Comprehend, Rekognition, etc.)
First Principle: AWS Managed AI Services provide pre-trained intelligence for common tasks, allowing businesses to integrate AI capabilities quickly without needing in-house ML expertise.
For many standard AI tasks, you don't need to build a custom model. You can use an AWS AI Service via a simple API call.
Key Services and Their Use Cases:
- For Vision:
- Amazon Rekognition: Analyze images and videos to detect objects, faces, text, and inappropriate content.
- Amazon Textract: Extract text and structured data (from forms and tables) from documents.
- For Language and Speech:
- Amazon Comprehend: Understand text to find sentiment, entities, and key phrases (NLP).
- Amazon Translate: Translate text between languages.
- Amazon Transcribe: Convert speech into text (ASR).
- Amazon Polly: Convert text into lifelike speech (TTS).
- For Conversational AI:
- Amazon Lex: Build chatbots and voice assistants.
- For Search and Enterprise Knowledge:
- Amazon Kendra: An intelligent enterprise search service that uses natural language processing to find answers within your company's documents, FAQs, and data sources. Unlike keyword search, Kendra understands the meaning of questions and returns precise answers ā not just a list of links.
- For Business Applications:
- Amazon Forecast: Predict future trends with time-series forecasting.
- Amazon Personalize: Build recommendation engines.
- Amazon Fraud Detector: Identify potentially fraudulent online activities.
Scenario: A company wants to add three features to its app: 1) Automatically moderate user-uploaded profile pictures for inappropriate content. 2) Translate user comments into English. 3) Provide an audio read-out of articles.
Reflection Question: Which three AWS AI Services would you map to these three requirements, and why is using these managed services more efficient than building custom models for these tasks?
š” Tip: When you see a common business problem like "understanding text" or "analyzing images," your first thought should be, "Is there a managed AWS AI Service for this?" before considering a custom build with SageMaker.
