Case Studies for the Professional Cloud Architect Exam¶

The current exam guide lists four case studies that questions may refer to. This note summarizes each in my own words, maps its requirements to Google Cloud design choices, and lists the traps. Read Google's originals first. They are short (one to two pages each) and the exam shows them next to the question.

📖 Altostrat Media case study (PDF) - media library, GKE, gen AI. 📖 Cymbal Retail case study (PDF) - online retailer, catalog enrichment, conversational commerce. 📖 EHR Healthcare case study (PDF) - SaaS health records, colocation exit, hybrid. 📖 KnightMotives Automotive case study (PDF) - car maker, AI, data monetization, EU rules. 📖 Professional Cloud Architect exam guide (PDF) - lists the cases and the exam sections.

Older material. Earlier exam guides used Mountkirk Games, TerramEarth, Helicopter Racing League and Dress4Win. They are not in the current guide. The six worked examples in solution-architecture-design.md are generic and still useful as design practice, but do not spend case-study time on the retired four.

How to use a case study in the exam¶

1. Read the constraints, not the story. Business requirements say why, technical requirements say what must be true. A wrong answer usually breaks one constraint.
2. Underline the words that filter answers: "99.9%", "legacy", "on-premises", "no budget for new equipment", "auditable", "EU", "reduce cost", "minimal operational overhead".
3. Prefer managed and the least change that meets the constraint. A case that says "legacy systems will not move" means hybrid connectivity, not migration.
4. Check the executive statement. It states the top priority (for example reliability and cost for Altostrat).
5. Questions test one decision at a time. Match the decision to a design pattern below.

1. Altostrat Media¶

Who: a media company with a large audio and video library. Wants gen AI for discovery, support and content insight.

Current state: content platform on GKE, media in Cloud Storage, BigQuery as the warehouse, Cloud Run functions for event tasks (transcoding, metadata, recommendations). Some legacy on-premises systems for ingestion and archive. Google Identity plus third-party IdPs. Cloud Monitoring and Prometheus, with alerts by email.

What they need, in short: - Reliable workflows across Google Cloud and on-premises, simple infrastructure management. - Lower storage cost for growing media, without losing availability. - Natural language, 24/7 support, summaries, metadata extraction (language and vision), harmful content detection, trend insight. - Modern CI/CD for containers, secure fast hybrid connectivity for ingestion, Kubernetes both on-premises and in cloud. - AI that is auditable and explainable.

Traps: storage cost questions want lifecycle and Autoclass, not deleting media. "Auditable and explained" is a governance requirement, so logging and review matter as much as the model. Do not pick a service that cannot run on-premises when the case demands Kubernetes on-premises.

2. Cymbal Retail¶

Who: a growing online retailer with a huge, messy product catalog.

Current state: mix of on-premises and cloud. MySQL, SQL Server, Redis, MongoDB. Kubernetes clusters. SFTP and batch ETL integrations. A custom catalog web app that searches by names and categories. An IVR with call-center agents who key in orders. Grafana, Nagios and Elastic for monitoring. Problems: manual work, data silos, hard to add new technology.

Three goals: automate catalog enrichment with gen AI (attributes, descriptions, images from supplier data), conversational product discovery, and modernize the stack.

Requirements to remember: reduce manual effort and errors, better discovery, higher conversion, lower call-center and data-center cost. Technical: attribute generation from titles, descriptions and images, image variations, natural language search, scale, human-in-the-loop review, data security and compliance.

Traps: gen AI output should not write straight to production, so the answer with a human review step usually wins. "Reduce data-center cost" points at migration and managed services, not new hardware. Keep PII out of prompts unless needed and protect it.

3. EHR Healthcare¶

Who: a SaaS provider of electronic health records to medical offices, hospitals and insurers, growing fast.

Current state: several colocation facilities, with a lease about to expire. Customer apps are web-based and many are containerized on Kubernetes. MySQL, SQL Server, Redis and MongoDB. Legacy file and API integrations with insurers stay on-premises with no plan to move them. Users in Active Directory. Open-source monitoring, with email alerts that people ignore.

Business needs: onboard insurers quickly, 99.9% availability, central visibility and proactive action, healthcare trend insight and predictions, lower latency, regulatory compliance, lower admin cost.

Technical needs: keep legacy insurer interfaces connected to both on-premises and cloud, consistent management of containerized apps, secure and fast on-premises link, consistent logging and retention, monitoring and alerting, several container environments, scale and provision environments dynamically, interfaces to ingest data from new providers.

Traps: the case says legacy systems will not move, so answers that migrate them fail. 99.9% does not require multi-region, and a simpler regional design is correct and cheaper. Alerting fixes need routing and ownership, not more email.

4. KnightMotives Automotive¶

Who: a car maker with electric, hybrid and combustion models. The newer in-vehicle experience is only on electric cars. Wants one modern experience across all models within five years and better dealer tooling.

Current state: mostly on-premises with some public cloud. Outdated mainframe supply chain, outdated ERP, many code bases and technical debt, siloed data, obsolete AI infrastructure. Dealers have no budget for new equipment. Weak connectivity at plants and in rural areas. Past data breaches.

Business needs: personalized relationship with the driver, consistent experience, better build-to-order with transparency, monetize data to fund technology, security paramount, EU data protection (especially for autonomous platforms), invest in autonomous driving, skills and communication.

Technical needs: consistent AI-enabled in-vehicle UX, reliable connectivity, plant-to-HQ network upgrades, hybrid cloud, gradual replacement of legacy, autonomous development and a simulation environment, data platform with strong security, security framework with incident response and training, improved online ordering, dealer tools, CRM.

Traps: EU rules point at residency, key control and auditability, so a US multi-region answer is wrong. "Dealers have no budget" rules out solutions that need on-site hardware. Autonomous vehicle compliance is a governance requirement, so log and audit.

Cross-case patterns¶

Practice¶

Write a one-page answer for each case: the three biggest constraints, the architecture in a Mermaid flowchart, the connectivity choice, the data choice, the security controls, and one thing you would NOT do and why. Then answer the case-study questions in the practice questions.