Black Friday sale: an extra 40% off every course until Monday, November 30.New Year sale: an extra 30% off every course until Friday, January 1.Memorial Day sale: an extra 25% off every course until Monday, May 31.SysAdmin Day sale: an extra 25% off every course until Sunday, August 1.Back to School sale: an extra 25% off every course until Sunday, August 22.Labor Day sale: an extra 25% off every course until Monday, September 6.Black Friday sale: an extra 40% off every course until Monday, November 29.New Year sale: an extra 30% off every course until Saturday, January 1.Choose your certification →

Copyright (c) 2026 MindMesh Academy. All rights reserved. This content is proprietary and may not be reproduced or distributed without permission.

3.6. Integrating Generative AI and Machine Learning into Architectures

💡 First Principle: A foundation model is another managed dependency behind an API. The architecture around it decides whether an AI feature is secure, affordable and operable: where it is called from, which data it can reach, which identities and guardrails govern it, how every call is logged, and whose bill each token lands on.

Scenario: A financial services company with 60 AWS accounts wants product teams to add generative AI features: a customer-facing assistant grounded in policy documents, an internal agent that opens tickets and updates CRM records, and a fraud model the data science team hosts on "Amazon SageMaker AI". Security requires central content controls and a full record of prompts and responses, finance wants per-team chargeback, and the platform team wants as little custom infrastructure as possible.

At the professional level you design the integration, not the model. The exam guide puts model training, fine-tuning and data science workflows out of scope. What it does test is whether you can choose between "Amazon Bedrock" and "SageMaker AI" endpoints, design retrieval-augmented generation (RAG) and agent architectures, apply guardrails and identity controls that the application cannot bypass, put a human approval step in front of risky AI actions, and run the result at a sensible cost with evaluation, monitoring and an audit trail.

Visual: Generative AI Reference Architecture

⚠️ Common Pitfall: Treating the model call as the whole design. Most AI incidents on AWS are ordinary architecture failures: a prompt log nobody enabled, a guardrail that only one team remembered to pass, an agent role with * permissions, or a GPU endpoint left running at 3 a.m.

Key Trade-Offs:
  • Managed vs. Self-Hosted Models: "Amazon Bedrock" gives serverless, per-token access to many models with no capacity to manage. "SageMaker AI" endpoints give full control of the model, container and instance type, but you size, scale and pay for the capacity.

Reflection Question: For the financial services scenario, which controls would you place centrally (in the management or a platform account) and which would you leave to each workload account, so that teams can ship AI features quickly without being able to switch off content filtering or the prompt audit trail?

Alvin Varughese
Written byAlvin Varughese
Founder•26 professional certifications earned