Foundation Model Integration, Data, and Compliance

This first AIP-C01 domain carries the most weight for a reason. AWS is testing whether you can choose the right GenAI application shape before you write code: direct FM calls, RAG, agents, workflow orchestration, customized models, multimodal ingestion, and the data boundary around each choice.

What this domain is really testing

Strong AIP-C01 answers start by matching the business requirement to a production architecture. Weak answers jump straight to “use a bigger model” or “fine-tune it” without checking retrieval quality, latency, compliance, grounding, lifecycle management, or data-processing requirements.

What AWS is explicitly testing

As of May 11, 2026, the current AWS Documentation exam guide splits this domain into six tasks:

• Task 1.1: analyze requirements and design GenAI solutions
• Task 1.2: select and configure FMs
• Task 1.3: implement data validation and processing pipelines for FM consumption
• Task 1.4: design and implement vector store solutions
• Task 1.5: design retrieval mechanisms for FM augmentation
• Task 1.6: implement prompt engineering strategies and governance for FM interactions

That split matters because Domain 1 is not only “RAG.” It covers architecture selection, model fit, data preparation, vector-store design, retrieval quality, and prompt governance as one connected design problem.

Current weight in the exam guide

AWS currently weights this domain at 31% of scored content on AIP-C01, making it the largest part of the exam. That is why model fit, vector stores, prompt design, and data boundaries show up repeatedly.

Official task compression

AWS splits Domain 1 into six tasks. This guide compresses those tasks into two learning paths so you can study by decision type instead of memorizing a long objective list.

Architecture and model fit covers Task 1.1 requirements analysis and Task 1.2 FM selection. Study 1.1 Architecture and Model Fit first when the scenario asks you to choose direct FM calls, RAG, agents, workflows, proof-of-concept scope, model routing, provider switching, resilience, or customization lifecycle.

Retrieval, prompts, and data boundaries covers Task 1.3 data processing, Task 1.4 vector stores, Task 1.5 retrieval mechanisms, and Task 1.6 prompt governance. Study 1.2 Retrieval, Prompts, and Data Boundaries first when the scenario asks about data validation, multimodal preparation, embeddings, chunking, metadata, hybrid search, reranking, query handling, prompt templates, prompt regression, or compliance-aware context assembly.

Architecture decision stack

Use this order when the exam asks how to design the first version of a GenAI application.

1. Name the knowledge problem. Decide whether the model already has enough general knowledge, whether the answer must use current/private sources, or whether the system must take action through tools.
2. Choose the application shape. Pick direct inference, RAG, knowledge bases, workflow orchestration, agentic tool use, model customization, or a hybrid pattern based on the failure mode, not on novelty.
3. Select the model path. Compare capability, latency, token cost, context window, regional availability, resilience, routing, provider switching, and lifecycle requirements.
4. Prepare the data path. Validate, classify, normalize, chunk, embed, index, synchronize, and authorize source material before it becomes prompt context.
5. Govern the prompt path. Version templates, define instruction hierarchy, validate variables, control output format, test regressions, and keep evidence for audit and rollback.

Work this domain in order

Start with 1.1 Architecture and Model Fit to separate direct FM use, RAG, customization, routing, resilience, and proof-of-concept decisions.

Then move to 1.2 Retrieval, Prompts, and Data Boundaries to tighten embeddings, vector-store choices, prompt operations, and compliance-aware data handling.

Fast routing inside this chapter

• Go to 1.1 Architecture and Model Fit when the scenario is about choosing direct inference, RAG, agent or workflow design, customization, model route, POC scope, fallback, dynamic provider/model selection, Cross-Region continuity, or graceful degradation.
• Go to 1.2 Retrieval, Prompts, and Data Boundaries when the scenario is about chunking, embeddings, vector-store architecture, metadata filters, hybrid search, reranking, data validation, multimodal or tabular preparation, source freshness, re-indexing, prompt templates, prompt versioning, prompt injection, output schema, or prompt regression.

High-yield decision rules

• If documents change often, prefer RAG, source synchronization, metadata, freshness checks, and retrieval evaluation before customization.
• If the model lacks task behavior repeatedly, compare stronger models, prompt governance, and customization lifecycle only after retrieval and context fit are understood.
• If the answer must cite internal source material, use retrieval with source IDs, access filters, and citation-aware prompt assembly.
• If the same app must switch models without redeploying code, use a routing layer with configuration-driven model selection and observability.
• If regulated or tenant data is involved, classify data, authorize before retrieval, encrypt, constrain logs, and preserve audit evidence.
• If a prompt update broke behavior, inspect prompt version, assembled request, variables, regression dataset, and rollback path.

Common traps

• using customization when grounding or retrieval would solve the problem faster
• treating prompt design as a substitute for bad source data
• forgetting that embeddings, chunks, and metadata are part of the security boundary
• choosing an architecture that works for a demo but has no clean rollback, evaluation, or audit path
• indexing sensitive data before classification, authorization, retention, and log controls are defined
• treating retrieved text as trusted instructions instead of evidence

In this section

• Architecture and Model Fit
  AIP-C01 lesson on GenAI architecture fit, model selection, Bedrock POCs, resilient routing, RAG, agents, and customization tradeoffs.
• Retrieval, Prompts, and Data Boundaries
  AIP-C01 lesson on RAG data preparation, vector-store design, retrieval tuning, prompt governance, and compliance-aware GenAI boundaries.