CloudArcade

Guided architecture tour · 6–8 min

Help an AI answer from your own documents — Bedrock and Knowledge Bases

Design an AI helper that looks up relevant documents before answering.

This tour teaches architecture decisions. Its services are not provisioned in the emulator.

Your goal

Design an AI helper that looks up relevant documents before answering.

Why this matters

A support helper needs answers from your product guide. You will prepare trustworthy sources, retrieve useful passages, and check that an answer is supported and allowed.

Words you will use

Retrieval means looking up useful information. RAG adds retrieved material to a model’s input. A citation points to a source the visitor can check.

How to proceed

Read each situation and choose an answer. The lesson explains the result. Try again if needed, then continue through all three decisions and finish the tour.

Play this architecture tour

1. Choose trustworthy material

Your help folder contains current guides, drafts, and private employee notes. The assistant should answer customer questions from approved material. Choose the source set.

What should it search?

Approved, current documents appropriate for the visitor.

A knowledge base is only as useful as its source material and access design. Remove stale or unsuitable content and plan updates. Retrieval should not expose information the visitor is not allowed to read.

The assistant has an appropriate source collection.

Approved sources → ingestion → searchable knowledge

2. Retrieve before answering

A customer asks about refunds. You want the model to use the relevant policy passages without treating this as model training. Choose the flow.

What happens first?

Retrieve relevant passages and include them with the question.

RAG retrieves material and supplies it as context for a response. It does not mean the model permanently learned that material. Useful retrieval helps, but it still does not guarantee a correct answer.

The answer can be informed by the refund policy.

Question → relevant passages → model response

3. Check support and boundaries

The answer looks confident, but a cited passage does not support its claim. Another question has no useful source at all. Choose the right quality rule.

What should the helper do?

Check support, show useful citations, and admit missing information.

Review whether the cited material actually supports the answer. Test missing information and access boundaries as well as easy questions. A clear “I could not find that” is better than an invented policy.

The helper has a checkable answer and a fallback for uncertainty.

Supported answer + source; otherwise acknowledge the gap

Practice, then check your understanding

In the interactive tour, make each decision before seeing the outcome. Wrong answers receive coaching, and replay gives you a fresh attempt without duplicating earned XP.

Read the official AWS documentation for service behavior and production details.

Continue with another AWS learning guide