ADVANCED · 8 MODULES · 64 LESSONS
Generative AI in Production on AWS
Run a generative AI workload on AWS like any serious system: infrastructure as code, guardrails, personal data protection, and token costs under control.
- For you if
- You already build with language models but have never run one in production on AWS.
- By the end
- You'll know how to define Bedrock and its guardrails in Terraform, restrict permissions, validate outputs, and calculate cost per token.
- You need
- The earlier AWS guides (or equivalent), Terraform, Python with pytest, and experience building with an LLM.

Curriculum
01Genai In Production Vs A Notebook
- Guide introduction: the eighth piece of the ecosystem
- A notebook and a production system are not the same problem
- Andes Cargo's AI workload: when the deterministic parser isn't enough
- The boundary with AI Engineering, said out loud
- Hands-on: inventory of what already exists
- Bedrock, 2026 landscape: models, families, and why the cheap path stays the default
- Hands-on: the honest attempt against Bedrock on LocalStack
- Project: Andes Cargo's AI workload map
02The Bedrock Cost Model
- Introduction: price per token, not per hour
- How Bedrock is billed: on-demand tiers, Batch, and Provisioned Throughput
- Bedrock quotas: the limit no bill shows
- Hands-on: Infracost, inherited
- Hands-on: the honest attempt of `infracost scan` on `bedrock.tf`
- Why Infracost can't solve this alone, and what can
- Hands-on: the cost-per-token calculator
- Project: Andes Cargo's `GENAI-COST-PROFILE.md`
03Infrastructure As Code For An Ai Endpoint
- Introduction: what Terraform knows how to declare without needing to invoke anything
- What Terraform resources exist for Bedrock
- Hands-on: the `modules/bedrock-guardrail/` module
- Hands-on: `BedrockManifestExtractorRole`, least privilege
- Hands-on: `terraform plan` of the complete AI infrastructure
- The exact limit: why the guardrail's `apply` doesn't run against LocalStack
- Hands-on: verifying what can actually be applied
- Project: Andes Cargo's AI infrastructure, declared
04Bedrock Guardrails And Defense In Depth
- Introduction: the managed guardrail and the custom one
- Bedrock Guardrails' six policies, explained
- Hands-on: declaring the complete guardrail in HCL
- Why a managed guardrail isn't enough alone
- Hands-on: the custom, deterministic PII scrubber
- Hands-on: the output schema validator
- The exact limit: why Bedrock Guardrails' real blocking can't be tested here
- Project: Andes Cargo's guardrails layer
05Securing The Ai Workload
- Introduction: the gate already exists, this workload enters it
- Least privilege for a model, not a service
- Hands-on: `bedrock-least-privilege.rego`, added to `policy/`
- Bedrock API keys exist: why this workload still uses an IAM role
- Hands-on: Trivy on the new Terraform
- Hands-on: signing the extractor's artifact with `cosign`
- STRIDE, revisited: the threats specific to an AI workload
- Project: Andes Cargo's security gate, extended
06Finops For Tokens
- Module introduction: the cost gate already exists, tokens are what's new
- Why a token is the ecosystem's most extreme *usage-based* resource
- Hands-on: `bedrock-budget.rego`, added to `cost-policy/`
- Hands-on: the per-token cost calculator, integrated into the gate
- Cost allocation tagging for the AI workload
- Hands-on: a volume change that triggers the budget
- On-demand vs. Provisioned Throughput: when to use each, with real numbers
- Project: Andes Cargo's *cost gate*, extended to tokens
07Observability Latency And Evals In Production
- Module introduction: the same vocabulary, new metrics
- SLIs for an AI workload: escalation, latency, guardrail block rate
- Hands-on: CloudWatch Logs and Metrics for the extractor
- Hands-on: the escalation rate metric
- What a production eval is, and why it isn't a unit test
- Hands-on: the smoke test harness with fixed manifests
- The exact limit: why real latency and quality can't be measured here
- Project: Andes Cargo's AI workload observability dashboard
08Capstone The Andes Cargo Genai Extractor
- Capstone introduction: the complete extractor, closed
- Architecture review: the full system, two paths
- End-to-end walkthrough: the deterministic, cheap path, executed
- End-to-end walkthrough: the AI escalation path, mixed and declared
- What this guide left representative
- The boundary with AI Engineering, closed
- What Andes Cargo still needs
- Final project: the Andes Cargo GenAI package as a deliverable
Reviews
What students say
These reviews are from enrolled students who completed at least 50% of the course. We moderate reviews only on content grounds (spam, offensive language, personal data), never for being critical or negative.
No approved reviews yet.
Be the first to share your experience!