Skip to content
Linear Horizon Training

Ready-made course · 2 days · Intermediate

AI Systems on AWS

A two-day course for cloud engineers, developers and architects: design and build AI systems on AWS, with architecture first and managed services in their proper place.

AWS offers a dense set of managed AI services. Teams that start from the product list often produce designs that are hard to move, hard to evaluate and expensive to run. This course starts from the system: model access, identity, retrieval, orchestration, events, security and cost. AWS services are then used as implementations of those parts.

You cover managed foundation models, identity and IAM, storage, serverless patterns, RAG and vector search, agents, event-driven systems, observability, deployment, security and cost. Service names are taught as current implementations, not as the architecture itself, so the design still holds when the console labels change.

The course is practical. You design and build against AWS primitives for model access, retrieval and an application path, then review the security and cost of what you built.

You leave able to design an AI system that happens to run on AWS, rather than an AWS diagram that happens to mention a model.

Who it is for
Cloud engineers, developers and architects working on AWS who will design or implement AI systems.
Before you start
Working knowledge of AWS identity, networking and at least one compute option (functions, containers or virtual machines). Familiarity with LLM applications is helpful.

You will be able to

  • Design an AI application architecture on AWS with explicit identity, data and model boundaries.
  • Integrate managed foundation models through a controlled access path rather than scattered credentials.
  • Implement RAG using storage, search and retrieval services appropriate to the corpus and tenancy model.
  • Place agents and event-driven workloads on queues and workflows with IAM-scoped tools.
  • Apply observability, security and cost controls that are native to the platform.
  • Deploy a working path and know which parts are portable if the model provider changes.

Modules

8 quarter-day modules. Each one can go into a custom course on its own.

  1. AI services on AWS

    • Bedrock, SageMaker and the managed AI services, and when each fits.
    • Accounts, regions and model availability.
    • Reference architecture for an AI application on AWS.
    • Cost model and quotas.
  2. Amazon Bedrock in practice

    • The Converse API, streaming and tool use.
    • Model access, inference profiles and prompt caching.
    • Bedrock Guardrails.
    • Lab: call several models through one interface.
  3. RAG on AWS

    • Bedrock Knowledge Bases: what they do and what they hide.
    • OpenSearch, Aurora pgvector and S3 Vectors.
    • Ingestion from S3 and other sources.
    • Lab: build a knowledge base and query it with citations.
  4. Agents on AWS

    • Bedrock Agents and AgentCore.
    • Tools as Lambda functions and APIs.
    • Memory, identity and observability in AgentCore.
    • Lab: deploy an agent with two tools.
  5. Serverless AI applications on AWS

    • Lambda, API Gateway and streaming responses.
    • Step Functions for multi-step AI workflows.
    • EventBridge and SQS for background work.
    • Lab: deploy an AI API with infrastructure as code.
  6. Securing AI workloads on AWS

    • IAM for model access and tools.
    • VPC endpoints, KMS and private networking.
    • Logging model invocations without leaking data.
    • Data residency and service availability by region.
  7. Operating AI on AWS

    • CloudWatch metrics, logs and traces for AI calls.
    • Quotas, throttling and provisioned throughput.
    • Cost allocation and budgets.
    • Evaluation jobs in Bedrock.
  8. Lab: an AI application on AWS, end to end

    • Put the pieces together: API, model, knowledge base and agent.
    • Deploy with CDK or Terraform.
    • Add guardrails, logging and an evaluation run.
    • Review the design against the Well-Architected generative AI lens.

Practical work

You build a small AWS-backed AI path: authenticated model access, a retrieval component, and either a synchronous application or an event-driven job. The closing review covers IAM, cost, observability and which interfaces should stay portable.

You take away

  • A working AWS implementation of a small AI system
  • Reference architectures mapped to current AWS services
  • IAM, cost and observability checklists
  • A portability note: which interfaces to keep stable

Combine it with

Contact