AWS GenAI services make up approximately 28% of AIF-C01 — the largest single domain. Core topics: Amazon Bedrock, SageMaker JumpStart, Amazon Q, PartyRock, and AI-specific hardware.
---
The Three-Layer Architecture
Think of building a house: raw material suppliers (infrastructure), construction companies (platform), and finished apartments (applications). AWS GenAI follows the same three layers.
| Layer | Analogy | AWS Service | Role | |-------|---------|-------------|------| | Infrastructure | Raw materials | Trainium, Inferentia | AI-specific chips for training/inference | | Platform | Construction | Amazon Bedrock, SageMaker | FM access, deployment, fine-tuning | | Application | Finished apartment | Amazon Q, PartyRock | Ready-to-use AI services |
!AWS GenAI three-layer architecture
Amazon Bedrock — The Most Tested Service
Bedrock is a fully managed service that gives access to multiple foundation models through a single API — like a music streaming service where you can choose from many artists.
Key Features
| Feature | What it does | |---------|-------------| | Model API | Call various FMs with a simple API | | Knowledge Bases | Automated RAG pipeline | | Agents | Multi-step task automation with external systems | | Guardrails | Content filtering, PII masking, topic denial | | Model Evaluation | Compare performance across multiple FMs | | Fine-tuning | Customize model for specific domain |
Pricing Models
| Model | Description | When to choose | |-------|-------------|---------------| | On-Demand | Pay per token used | Unpredictable or low traffic | | Provisioned Throughput | Reserve capacity, fixed cost | High, stable traffic | | Batch Inference | Process large batches together | Non-real-time workloads |
---
Bedrock vs SageMaker JumpStart
| Comparison | Amazon Bedrock | SageMaker JumpStart | |-----------|----------------|---------------------| | Model management | AWS/partners manage | Deploy to your own infrastructure | | Customization | Limited fine-tuning | Full customization | | Complexity | Low (fully managed) | High (manage infrastructure) | | Primary user | App developers | ML engineers |
Exam tip: "Deploy to own infrastructure, fine-grained control" → SageMaker JumpStart / "Simple API to call FM" → Bedrock
---
Amazon Q
Amazon Q Business: AI assistant that searches internal company data (40+ connectors for Salesforce, SharePoint, S3, Confluence, etc.) Amazon Q Developer: Coding assistant integrated with IDEs — code completion, security scanning, test generation
---
Hardware
| Chip | Purpose | Key benefit | |------|---------|-------------| | AWS Trainium | ML model training | Up to 50% cost reduction vs GPU | | AWS Inferentia | Model inference | Low-cost, high-throughput inference |
---
Exam Key Points
"Single API for multiple foundation models" — Amazon Bedrock "Automated RAG pipeline" — Bedrock Knowledge Bases "Content filtering, PII masking" — Bedrock Guardrails "Deploy FM to own infrastructure" — SageMaker JumpStart "Internal data AI assistant" — Amazon Q Business "Code generation and security scanning" — Amazon Q Developer "No-code GenAI app builder" — PartyRock "Training chip" — Trainium / "Inference chip" — Inferentia