Build Exercises

Domains 1–5 include task-level build exercises from the study guide.

Domain 1: Foundation Model Integration, Data Management, and Compliance

6 exercises

1.1Intermediate45 minutes

1.1 — Build a GenAI Architecture Decision Prototype

What you will learn

Map business and technical requirements into a GenAI architecture.
Justify FM choice, retrieval strategy, security, monitoring, and fallback.
Explain why the architecture satisfies constraints rather than maximizing sophistication.

Open Domain 1 lessons First skill in task 1.1

1.2Intermediate40 minutes

1.2 — Build a Model Router with Failover

What you will learn

Model abstraction and configuration-driven routing.
Dynamic provider/model switching without application rewrites.
Fallback, resilience, and observability for FM calls.

Open Domain 1 lessons First skill in task 1.2

1.3Intermediate40 minutes

1.3 — Build a Multimodal Data Preparation Pipeline

What you will learn

Validate and route text, image, audio, and tabular inputs.
Normalize outputs into model-compatible formats.
Record quality metrics before FM inference.

Open Domain 1 lessons First skill in task 1.3

1.4Advanced50 minutes

1.4 — Build a Document-to-Vector Knowledge Pipeline

What you will learn

Ingest documents from S3 through chunking and embedding into a vector store.
Attach document IDs, timestamps, domain, and access metadata.
Support incremental updates and failure handling.

Open Domain 1 lessons First skill in task 1.4

1.5Advanced50 minutes

1.5 — Build a Hybrid RAG Retrieval Layer

What you will learn

Combine vector and keyword retrieval with reranking.
Apply metadata filters before returning context to the FM.
Use query decomposition for complex multi-part questions.

Open Domain 1 lessons First skill in task 1.5

1.6Advanced45 minutes

1.6 — Build a Governed Prompt Evaluation Pipeline

What you will learn

Version prompts and gate them with evaluation before production.
Validate outputs and record audit/metrics signals.
Support rollback to a previous prompt version.

Open Domain 1 lessons First skill in task 1.6

Domain 2: Implementation and Integration

5 exercises

2.1Advanced50 minutes

2.1 — Build a Safeguarded Multi-Step Agent

What you will learn

Separate state from long-term memory in an agent workflow.
Use tools through a standardized MCP-style interface.
Add stopping conditions, IAM boundaries, and human approval.

Open Domain 2 lessons First skill in task 2.1

2.2Intermediate40 minutes

2.2 — Choose and Justify an FM Deployment Path

What you will learn

Map application requirements to Lambda, Bedrock Provisioned Throughput, or SageMaker AI.
Account for LLM-specific concerns: memory, GPU, tokens, and model loading.
Apply cascading so simple requests use smaller models.

Open Domain 2 lessons First skill in task 2.2

2.3Intermediate45 minutes

2.3 — Integrate GenAI into an Existing Enterprise App

What you will learn

Add GenAI through APIs, webhooks, and event-driven paths without rewriting the core app.
Apply identity federation, RBAC, and least privilege.
Introduce a GenAI gateway and CI/CD promotion path.

Open Domain 2 lessons First skill in task 2.3

2.4Advanced45 minutes

2.4 — Build a Resilient FM API Layer

What you will learn

Support sync, async, and streaming FM interactions.
Handle retries, rate limits, fallback, and tracing.
Route requests statically or dynamically across models.

Open Domain 2 lessons First skill in task 2.4

2.5Intermediate40 minutes

2.5 — Ship a Developer-Ready GenAI Application Path

What you will learn

Expose FM capabilities through stable application interfaces.
Use Prompt Flows and business-system enhancements thoughtfully.
Troubleshoot with Logs Insights, X-Ray, and developer assistance tools.

Open Domain 2 lessons First skill in task 2.5

Domain 3: AI Safety, Security, and Governance

4 exercises

3.1Advanced55 minutes

3.1 — Production-Grade Security & Governance Pipeline

What you will learn

Configure Bedrock Guardrails for denied topics, PII actions, and grounding.
Isolate Bedrock calls with PrivateLink and least-privilege endpoint policies.
Alert compliance automatically when Guardrails intervene.

Open Domain 3 lessons First skill in task 3.1

3.2Intermediate35 minutes

3.2 — PII Discovery Before RAG Indexing

What you will learn

Discover sensitive data in S3 before embeddings.
Choose ANONYMIZE vs BLOCK for interaction-time PII.
Apply retention rules for chat transcripts.

Open Domain 3 lessons First skill in task 3.2

3.3Intermediate40 minutes

3.3 — Governance Evidence Pack for Auditors

What you will learn

Document models with Model Cards.
Trace data lineage and invocations.
Map each audit requirement to an AWS control.

Open Domain 3 lessons First skill in task 3.3

3.4Intermediate35 minutes

3.4 — Responsible AI Evaluation & Oversight Loop

What you will learn

Run fairness evaluations before production.
Expose reasoning/attribution for transparency.
Escalate consequential interventions to humans.

Open Domain 3 lessons First skill in task 3.4

Domain 4: Operational Efficiency and Optimization for GenAI Applications

3 exercises

4.1Advanced55 minutes

4.1 — Cost-Optimized & Latency-Tuned Document Summarization

What you will learn

Combine semantic cache, model cascading, and Prompt Caching.
Stream tokens for better perceived latency.
Alarm on token-spend regressions.

Open Domain 4 lessons First skill in task 4.1

4.2Intermediate35 minutes

4.2 — Tune Retrieval Latency for a RAG Path

What you will learn

Apply metadata pre-filters before k-NN search.
Tune OpenSearch/HNSW parameters.
Measure retrieval latency separately from FM latency.

Open Domain 4 lessons First skill in task 4.2

4.3Intermediate40 minutes

4.3 — FinOps Dashboard & Token Alarm Pack

What you will learn

Enable invocation logging for token forensics.
Dashboard tokens, latency, throttles, and cascade mix.
Alert when spend or throttling regresses.

Open Domain 4 lessons First skill in task 4.3

Domain 5: Testing, Validation, and Troubleshooting

2 exercises

5.1Advanced50 minutes

5.1 — Automated RAG Regression Pipeline & Diagnostic Suite

What you will learn

Detect stopReason max_tokens truncations in Converse metadata.
Score faithfulness with an LLM-as-a-Judge and mitigate position bias.
Gate deployments on FaithfulnessScore and TruncationCount.

Open Domain 5 lessons First skill in task 5.1

5.2Intermediate35 minutes

5.2 — Troubleshoot Truncation, Permissions, and Stale RAG

What you will learn

Separate max_tokens output truncation from ValidationException input overflow.
Fix agent AccessDeniedException with lambda:InvokeFunction.
Recover missing S3 docs via StartIngestionJob and Hybrid Search for exact codes.

Open Domain 5 lessons First skill in task 5.2