Skip to content

Module 1 · Architecture

AI architecture designer

Describe the system you want to build. We recommend a pattern (simple RAG, agentic RAG, GraphRAG, text-to-SQL or an agent), explain why, and lay out every layer, from gateway and guardrails to logging and cost monitoring, with the top tools compared on pros, cons and cost-effectiveness. Azure, AWS, Google Cloud and SAP BTP are covered.

  • About 3 minutes
  • RAG / agent pattern
  • Full-stack architecture
  • Top-10 tool comparisons
  • Residual risk and cost

Pre-filled with a common case: an internal knowledge assistant for 30,000 employees with personal data in the documents. Change anything that doesn't match.

1/4 Use case

What does the AI system decide or support?
What should the system mainly do?

2/4 Knowledge

What does the system need to know that the model doesn't?
How fresh must that knowledge be?
Does output need a specialized style or format?

3/4 Risk

What is the most sensitive data the model can see?
Where are users or affected people located?
Who can send input to the system?
How much autonomy does the agent have?

4/4 Platform

How fast must answers arrive?
Where do you run today?
Methodology and official sources

The designer is rule-based and deterministic: the same answers always produce the same design. The pattern follows from the shape of your knowledge and whether the system must act. Controls come from the OWASP Top 10 for LLM Applications, the NIST AI RMF and NIST AI 600-1, and your EU AI Act classification. Each layer cites its source.

Each layer compares up to 10 tools, best fit for the platform you chose first, with pros, cons and a link to the maintainer's documentation. Tools are options, not endorsements. Cost-effectiveness is a relative indication (open source or already-licensed, pay per use, or premium enterprise licence), not a price: confirm on each vendor's pricing page. The residual-risk score runs the audit engine on the design with its recommended controls in place.

↑↓ navigate↵ openEsc close