Shuichiro Ogawa
日本語

Notes · updated 2026-08-07

Overview

The question of this note is whether tools exist that help learners acquire correct system architecture — by pointing out missing elements in a design, or by providing exercises on design decisions. The subject is learning system design (architecture), not implementation coding. The 23 collected data points are organized into four types by how the learner interacts with them.

  • Review type: submit a design (questionnaire answers, IaC, a diagram) and get back missing elements and risks.
  • Exercise-with-feedback type: work through design problems and get step-by-step critique or mock-interview feedback.
  • Reference type: read exemplars of correct configurations (reference architectures, design patterns, selection guides).
  • Notation type: notations and modeling tools for learning to write designs down.

Wiring-and-load simulators were covered in the addendum of the adjacent note (Pre-Production Validation and Learning for System Architecture). This note complements it: learning to build correctly, rather than learning by breaking.

Review Type: Tools That Point Out What a Design Lacks

Mechanisms that mechanically point out missing design elements already exist as cloud vendors’ assessment tools.

The AWS Well-Architected Tool is a free tool in the AWS console: register a workload, answer multiple-choice questions across six pillars (operational excellence, security, reliability, performance, cost, sustainability), and receive a list of High Risk Issues and an improvement plan. The Lens Catalog adds perspectives such as Serverless and Generative AI, and instructors can define and distribute their own questions and improvement plans as a Custom Lens. Because each question presents the best practice and the reason for it, a student who registers a mock workload can see what their design lacks, expressed in the six-pillar vocabulary.

The Azure Well-Architected Review is the same shape of self-assessment: roughly 60 multiple-choice questions return per-category recommendations across five pillars (about 60 minutes). The official documentation recommends using it at the greenfield (initial design) stage, which fits the learning loop of entering design decisions and getting back the gaps. Google Cloud, by contrast, has the Architecture Framework documentation but no questionnaire-style assessment tool comparable to the other two vendors could be found.

For submitting a diagram directly, the AWS Well-Architected IaC Analyzer (aws-samples) analyzes CloudFormation or Terraform templates and also architecture diagrams (PNG/JPEG) with Claude on Amazon Bedrock, and outputs the gaps against Well-Architected best practices, prioritized by risk and complexity. It is the most direct match for “submit a diagram, get back what is missing,” but it is a sample project (explicitly non-production), requiring deployment into one’s own AWS environment and incurring Bedrock costs. Azure Advisor, although it also returns recommendations, analyzes deployed resources’ telemetry after the fact, so it does not suit design-stage learning.

Exercise-with-Feedback Type: Services That Critique a Learner’s Design

Services in which a learner produces a design and receives critique are concentrated in the system-design interview-practice market.

The only one confirmed to evaluate a drawn design diagram is HelloInterview. As the learner draws a design on a whiteboard while explaining it, an AI reads the drawing and returns instant, rubric-based feedback per step (34 system design problems; a free tier exists). The founder (a former engineering manager at Amazon and Meta) discloses the implementation internals on the blog (GPT-4o at the core of feedback generation, roughly $10,000/month in AI costs, CRDT-based whiteboard sync), so the first-person testimony is unusually transparent. The claim that the rubrics were “tuned by FAANG interviewers” was separated as a marketing claim because the process is undisclosed.

The others take different forms. interviewing.io offers anonymous mock interviews with practicing senior-plus engineers (paid) for human review, plus an AI interviewer covering system design with 200+ free questions. Exponent offers peer mock interviews (five per month free) with AI feedback added in 2025. Educative’s Grokking course includes eight AI mock interviews, but whether they evaluate the diagram itself could not be confirmed. The well-known reading materials ByteByteGo and DesignGurus have no design-submission feedback feature (the former added coding exercises only).

Reference Type: Exemplars of Correct Configurations

All three vendors publish free documentation portals as references for what a correct configuration is. The AWS Architecture Center and Solutions Library provide reference architectures, validated solutions, and Decision Guides for service selection. The Azure Architecture Center adds decision-tree-style technology selection guides and an anti-pattern catalog to its design patterns, letting learners study why a configuration is chosen. The Google Cloud Architecture Center provides a comparable body of reference architectures. None of these actively point out gaps, so they pair naturally with the review type.

Notation Type: Notations and Tools for Writing Designs

Learning design requires being able to write designs down. The de facto standard at this layer is the C4 model (describing architecture in four levels: Context, Container, Component, Code; created by Simon Brown), with Structurizr as its reference implementation generating diagrams from a single DSL. Structurizr Lite is free locally, and the cloud version is free for academic projects (on application), so it can be used in university courses as is. For the enterprise-architecture notation ArchiMate there is the free open-source tool Archi, whose official site explicitly mentions use by universities and students. The C4-based collaborative tool IcePanel also has a free tier. All of these are description and visualization tools with no automated judgment of design quality. As an evaluation methodology there is SEI’s ATAM (a stakeholder workshop method for evaluating quality-attribute tradeoffs), but no official software tool supporting it could be found.

Assembling This for Student Education

The four types are not mutually exclusive; within the scope of this survey, the following sequence composes naturally. Read the reference type (Azure’s decision-tree guides, AWS Decision Guides) to learn why components are chosen; have students write their own designs in the notation type (C4 with Structurizr Lite, both free); have them answer the review type’s six-pillar questions (AWS Well-Architected Tool, free) to find the gaps; and let them experience diagram-level critique in the exercise type (HelloInterview’s free tier). Instructors who want their own grading criteria can encode them as a Well-Architected Custom Lens. To then show how a design behaves under load, the wiring simulators organized in the addendum of the adjacent note are the natural next step.

Gaps

  • Products confirmed to have an AI evaluate a learner’s design diagram and point out gaps are limited to HelloInterview and the AWS IaC Analyzer (a sample); as dedicated educational products, the market is thin.
  • Google Cloud has no questionnaire-style assessment tool, so the review type depends on AWS and Azure.
  • No tooling for architecture evaluation methods such as ATAM could be confirmed.
  • No independent survey shows educational effectiveness (whether design skill actually improves) for any of these services; effectiveness research belongs to the peer-reviewed learning sciences (academic mode).
  • No methodologically explicit survey was found on university adoption of the modeling tools (Structurizr, Archi).

References

All accessed 2026-08-07. For ledger details (methodology, position judgments, all verification flags), see source/review/system-design-learning-tools/industry.md.

Vendor official (T1v)

Third party (T3)

← All Notes · Home