Business Problem
Customer uploads voice.
Need
Speech
↓
Translate
↓
Summarize
↓
Generate response
↓
Speech output
Architecture
Voice
S3
Lambda
Transcribe
Bedrock
Translate
Polly
Customer
Flow
Audio uploaded
↓
Amazon Transcribe
↓
Text
↓
Bedrock summarizes
↓
Translate
↓
Amazon Polly speaks
↓
Return audio
Services
S3
Lambda
Transcribe
Bedrock
Translate
Polly
Why Not Let LLM Process Audio?
Foundation models are text-based unless using multimodal models.
AWS speech services provide better accuracy and lower cost.
Decision Matrix (Exam Favorite)
Requirement Best AWS Service Why
Store documents S3 Durable, scalable object storage
OCR Textract Structured text extraction
Semantic search Bedrock Knowledge Bases Managed RAG
Generate embeddings Bedrock Managed embedding models
Business logic Lambda Serverless orchestration
REST API API Gateway Secure API front door
User authentication IAM / Amazon Cognito Secure access control
Structured data DynamoDB Low-latency NoSQL storage
Logs CloudWatch Monitoring and troubleshooting
Voice to text Transcribe Automatic speech recognition
Text to speech Polly Natural speech synthesis
Translation Translate Neural machine translation
Architecture Selection Cheat Sheet
Scenario AWS Architecture Pattern
Chat with enterprise documents S3 → Knowledge Base → Bedrock → Lambda
Customer support with live order data API Gateway → Lambda → Database → Bedrock
Invoice processing S3 → Lambda → Textract → Bedrock → DynamoDB
Personalized recommendations API Gateway → Lambda → DynamoDB → Bedrock
Voice assistant S3 → Transcribe → Bedrock → Translate → Polly
Key exam takeaway
A recurring pattern in AWS AI architectures is to let managed AWS services perform specialized tasks (for example, Textract for OCR, Transcribe for speech-to-text, or Knowledge Bases for retrieval) and use AWS Lambda to orchestrate workflows, retrieve business data, and enforce authorization. Amazon Bedrock is then used primarily for generative AI tasks such as summarization, question answering, content generation, or reasoning, rather than directly accessing enterprise systems or replacing specialized AI services. This separation of responsibilities is a common design principle tested in AWS AI Professional certification scenarios.
No comments:
Post a Comment