Скачать или смотреть AWS Machine Learning Associate Exam Walkthrough 23 AWS Kendra

AWS Machine Learning Associate Exam Walkthrough 23 AWS Kendra - September 21
VIEW RECORDING: https://fathom.video/share/kbMtcU79aT...
Meeting Purpose

To provide a comprehensive overview of Amazon Kendra and its integration with Amazon Augmented AI (A2I) for intelligent enterprise search.

Key Takeaways

Amazon Kendra is an ML-powered enterprise search service that understands natural language queries, extracts precise answers, and integrates with diverse data sources.
Kendra's intelligence stems from NLP, answer extraction, incremental learning, and relevance tuning capabilities.
A2I provides human-in-the-loop validation for Kendra, crucial for compliance and high-accuracy scenarios.
Pricing: Developer Edition (~$810/mo for 1k docs, 10k queries), Enterprise Edition (pay-per-use, millions of docs); A2I costs vary ($0.0012-$1+ per task).

Topics

Kendra's Intelligent Search Capabilities

Evolves beyond keyword matching to understand context, intent, and meaning
Uses NLP to interpret query intent (e.g., "Where is IT Support Desk?" → "first floor")
Indexes diverse document formats: PDFs, Word, PowerPoint, HTML, plain text, FAQ pairs
Connects to various data sources: S3, SharePoint Online, Confluence, Salesforce, RDS
Custom API connectors available for specialized integration needs

Machine Learning Features

Natural language understanding for conversational queries
Answer extraction pinpoints exact information within context
Incremental learning improves accuracy through user interactions and click patterns
Relevance tuning allows admin control over ranking factors (freshness, weights, permissions, metadata)

Console Walkthrough

Index creation process: naming, IAM role selection, edition choice (Developer/Enterprise/Gen AI)
Access control options: token-based user access, Identity Center integration
Data source configuration: S3 bucket setup, sync schedules (hourly/daily/weekly)
Testing via console search tab with natural language queries
Security optimization: document attributes, business importance, IAM policies, Active Directory integration

Amazon Augmented AI (A2I) Integration

Provides human-in-the-loop capability for ML predictions requiring oversight
Use cases: compliance mandates, accuracy-critical workloads, low confidence thresholds
Integrates with Amazon Rekognition, Textract, and custom SageMaker models
Setup in SageMaker console: create human review workflows, define routing conditions
Workforce options: Mechanical Turk, AWS Marketplace vendors, private teams
Results monitoring and S3 storage for audit trails and model improvement

Implementation Strategies

Start with high-value use cases: IT docs, HR policies, compliance materials
Integrate with other AWS services: Lex chatbots, QuickSight dashboards, custom SDK apps
Ideal for scenarios like legal firms searching case files or healthcare orgs navigating patient records

Next Steps

Explore Kendra's Developer Edition for proof-of-concept work
Identify high-value use cases within the organization for initial deployment
Evaluate A2I integration needs based on compliance and accuracy requirements
Consider integration opportunities with existing AWS services (Lex, QuickSight)
Stay tuned for the next session on Amazon's hardware for AI on the AWS platform