Skip to Content
Cloud Engineering Academy
17. Daily-Work Playbooks & Career

Interview preparation: cloud engineer

Prepare for a cloud engineer interview by turning your .NET and Azure experience into AWS stories, rehearsing a core bank of IAM, networking, operations and troubleshooting answers, and practising how to think aloud when you do not know the answer.

intermediate⏱ 2 h 🟒 FreeSAA-C03SOA-C03SCS-C03

Assumes: .NET 10 LTS Β· AWS CLI v2 Β· Region us-east-1. Verified against official sources on 2026-11-18.

Learning objectives

By the end of this lesson you can:

  • Describe what a cloud engineer interview loop tests and which part you are being scored on at each stage.
  • Build a story bank of six to eight real stories from your .NET and Azure work, plus the playbooks and labs from this course, in a format you can adapt in 90 seconds.
  • Answer core questions on IAM, networking, operations, security, infrastructure as code and cost with a short, correct, banking-aware answer.
  • Walk through a troubleshooting scenario aloud using the playbook method: scope, evidence, hypotheses, test, fix, prevent.
  • Handle "I have not done that in production" honestly, and show how you would find out.
  • Ask the interviewer questions that tell you whether the team is one you want to join.

Prerequisites

Why this matters in banking

You can run a flawless deployment on the day and still lose the job in an hour of conversation. Hiring managers at banks are not trying to find the candidate who has memorised the most services. They are looking for a person they would trust on the on-call rota at 02:00 during an incident that touches customer money. That trust comes from signs that you reason from evidence, you know the limits of what you know, you respect change control, and you communicate clearly to people who are not engineers.

You also have an unusual advantage. A .NET banking developer moving to AWS already understands transactions, audit trails, segregation of duties and release approvals. Many candidates who know AWS better have never lived with those. Your job in the interview is to show you have translated that experience, not that you have started from zero.

Market and salary claims

This lesson makes no claims about hiring volumes, salaries or which job titles pay what. Those change by country, year and employer. If you need a number, look at current job postings and salary surveys for your market, treat any single source with caution, and ask recruiters directly. ⚠ VERIFY anything you read, including anything that sounds authoritative.

The idea in plain language

A cloud engineer interview is usually four conversations in a trench coat:

StageWhat it is really testingHow you win it
Recruiter or HR screenFit, motivation, logistics, communicationA 60-second story: who you are, why cloud, why this bank, what you have built
Technical screenBreadth of fundamentals and honesty about gapsShort, correct answers; saying "I would check X" rather than guessing
Hands-on or scenario roundHow you diagnose and decideThinking aloud with a method, asking scoping questions, naming trade-offs
Behavioural / hiring managerJudgement, ownership, collaboration, change controlSpecific stories with a clear role, a result and a lesson

Some loops add a take-home task, a small design discussion (see interview preparation for solutions architects) or a live coding step (see interview preparation for .NET cloud developers). The names vary; the four things being tested do not.

Analogy: An interview is a flight check ride, not a written exam. The examiner cares less whether you can recite the manual and more whether you check the instruments, state what you are doing, and take the safe option when something looks wrong.

The three answer shapes

Most questions fit one of three shapes. Practise each until the shape is automatic.

1. The definition-and-why shape ("What is a security group?"): One-sentence definition, one-sentence why it matters, one concrete banking example, one limit or gotcha. Four sentences, about 30 seconds.

2. The scenario shape ("Calls to the partner API are timing out"): Scope and safe mode, evidence, hypotheses ranked, test one at a time, fix, prevent, record. This is the same shape as every playbook in this module. Say the steps out loud.

3. The story shape ("Tell me about a time you…"): Situation (one sentence), constraint (what made it hard), action (what you did), result (a number or a fact), lesson (what you changed). About two minutes. If you cannot name your own action, the story is not ready.

What problem the interview solves

For the employer, it is a risk reduction exercise: find out whether you can work safely and independently in a production environment with limited supervision. For you, it is a chance to find out whether the team's engineering practice matches what you want to learn. Both sides have honest limits:

  • A 45-minute conversation cannot measure depth. Interviewers over-weight confidence and fluency, and under-weight careful reasoning. You can use that by being clear and structured.
  • Interviewers differ. Some ask trivia ("what port does RDP use?"); some run scenarios. Prepare for both, and do not take an odd question personally.
  • You will be asked things you do not know. How you respond is often scored more than the question itself.

Microsoft / Azure comparison

How to translate what you already do into the vocabulary an AWS interviewer expects, without pretending you did AWS work you did not do.

What you did (Microsoft/Azure/.NET)How to say it for AWSBe careful to say
Deployed to Azure App Service or AKS"Container-based hosting, so on AWS I would use ECS on Fargate; I have built that in the course labs"What was real work and what was lab work
Used Azure Key Vault and managed identities"Same principles: no secrets in code, short-lived credentials. On AWS that is IAM roles plus Secrets Manager and KMS"You have not used the AWS services in production, if that is the case
Application Insights and Log Analytics"I diagnose from structured logs, traces and metrics; on AWS that is CloudWatch Logs Insights and X-Ray or OpenTelemetry"Which tools you have actually queried
Azure DevOps pipelines with approvals"Pipelines with environment approvals and audit; on AWS the same idea with GitHub Actions and OIDC or CodePipeline"The approval and evidence process you personally followed
Service Bus with dead-letter queues"Queues with retries and a DLQ; SQS with maxReceiveCount and redrive"The idempotency design you used
RBAC and Privileged Identity Management"Least privilege and just-in-time elevation; IAM roles, Identity Center permission sets and break-glass access"That the model differs: AWS has no direct PIM equivalent ⚠ VERIFY current options
Azure Policy and management groups"Guardrails at organisation level; SCPs and Config rules in AWS Organizations"Which guardrails you owned

Rule for every row: name the principle, name the AWS service, and be exact about your own experience. An interviewer forgives a gap. They do not forgive a bluff that collapses at the second follow-up.

Architecture

How the loop fits together

Thinking aloud in a scenario

Say the numbers aloud as you go. Interviewers often write down the steps they hear.

Step-by-step implementation

Build your story bank

Create a file with six to eight stories. For each, fill one block (a template follows in the Complete code section). Cover at least: an incident you handled, a change you made safely, a disagreement, a mistake you made, something you automated, a time you learned a new technology quickly and a time you improved cost, security or reliability.

Where you lack workplace examples for an AWS topic, use the playbooks and labs from this course, labelled clearly as practice ("in a lab, I reproduced…"). A reproduced and documented incident is still a real skill.

Prepare a 60-second opening

Structure: where you are now, what you have done that is relevant, why cloud and why this team. Write it, time it, cut it to 60 seconds. For example: "I have eight years building .NET payment services in a regulated bank, including ownership of release evidence and incident response. Over the last year I have been moving that experience onto AWS: I built a secure API on Fargate with Cognito and KMS, and I wrote runbooks for IAM and KMS denials. I want a platform role where I can apply cloud skills inside strong change control." Fictional but realistic; replace with yours.

Drill the core question bank

Use the question bank below. For each question, say your answer aloud in under a minute, then check it against the model answer. Drill the ones you got wrong again two days later.

Rehearse troubleshooting aloud

Pick one playbook a day. Without notes, explain it as if the interviewer had given you the symptom: scope, evidence, hypotheses, tests, fix, prevent. Record yourself or use a friend. You are checking that you give a method, not a list of random ideas.

Prepare your questions for them

Ask about on-call, incident reviews, change process, how they handle access to production, how they decide what to automate, and how engineers learn. Their answers tell you whether you want to work there.

Do a mock interview and fix one thing

Run a full mock with a person or the AI assistant (Lab 17.4). Pick one improvement only: shorter answers, more specific results, or naming trade-offs. Repeat.

Core question bank with model answers

Short answers are better than long ones. Offer to go deeper.

IAM and identity

  • What is the difference between an IAM user, a group and a role? A user is a long-term identity with credentials, a group is a collection of users for attaching permissions, and a role is an identity you assume for short-lived credentials. In a modern account humans use Identity Center and workloads use roles; long-term access keys are a last resort.
  • How does AWS decide whether a request is allowed? Start with an implicit deny. An explicit deny anywhere wins. Otherwise the request needs an allow in an applicable policy, and it must also pass any organisation guardrails (SCPs, RCPs), permission boundaries and session policies. For same-account access, an identity or resource policy allow can be enough; for cross-account access you need both sides.
  • What is the first thing you check on an AccessDenied? Read the error: which principal, action and resource. Then find the layer that denied it with CloudTrail and the policy simulator, rather than adding broad permissions to make it go away. See Policy evaluation logic for the full evaluation chain.
  • What is least privilege in practice? Start from what the workload does, grant only those actions on those resources with conditions, then use last-accessed data and Access Analyzer to remove what it does not use.
  • Why use roles for an ECS task rather than access keys? The task receives short-lived credentials automatically, there is nothing to leak or rotate by hand, and CloudTrail records the role session.

Networking

  • Security group or network ACL: which and why? Security groups are stateful, attached to network interfaces, allow rules only. NACLs are stateless, attached to subnets, ordered allow and deny rules. Use security groups as the main control; NACLs as a coarse extra layer. See Security groups versus NACLs.
  • How does a private subnet reach the internet? Through a NAT gateway in a public subnet with a route from the private subnet's route table. The gateway has an Elastic IP, charges per hour and per GB processed, and is a common source of cost surprises.
  • How do you reach S3 from a private subnet without the internet? A gateway VPC endpoint for S3 (and DynamoDB), which is free and keeps traffic off the NAT gateway. Interface endpoints (PrivateLink) cover most other services and have hourly and per-GB charges.
  • A service in one subnet cannot reach another. How do you debug? Check security group rules in both directions, NACLs, route tables, DNS resolution, then use VPC Flow Logs and Reachability Analyzer. Say which you would try first and why.
  • What is the difference between a load balancer idle timeout and a target timeout? The idle timeout closes a connection with no traffic between the client and the load balancer; target response problems show as 502, 503 or 504. Which hop returned the error tells you where to look.

Operations and monitoring

  • What would you alarm on for a payments API? User-visible symptoms first: error rate, latency percentiles and queue age, then saturation signals. Alarm on symptoms, not on every cause, and attach a runbook to each alarm.
  • CloudWatch metrics, logs and traces: how do they differ? Metrics are numbers over time, logs are events, traces follow one request across services. You use metrics to notice, logs to explain and traces to locate.
  • How do you find out who changed something? CloudTrail: management events are on for 90 days in event history; for longer retention you use a trail to S3. Data events must be enabled.
  • How do you handle an alarm at 02:00? Acknowledge, scope the impact, check recent changes, follow the runbook, stabilise (roll back or scale), communicate status, then record the timeline and write the follow-ups. Do not fix and forget.
  • How do you avoid alert fatigue? Fewer, better alarms; page only on customer impact; review every page after the incident; delete alerts that never lead to action.

Infrastructure as code and delivery

  • Why infrastructure as code in a bank? Repeatability, review, drift detection and an audit trail of who changed what and why. A reviewed pull request is better evidence than a console session.
  • CDK, CloudFormation or Terraform? CloudFormation is the engine for CDK; CDK lets you use C#; Terraform is multi-cloud and widely used. Say what the team uses, and what trade-offs you see: language fit, state handling, ecosystem and policy checks.
  • How do you deploy safely? Small changes, automated tests, scans, approval for production, progressive rollout (blue/green or canary), a rehearsed rollback and evidence kept per release.
  • What is drift? The live resource no longer matches the code, usually because of a manual change. Detect it, then either update the code or revert the resource through the pipeline.

Security and compliance

  • Where do you start securing an AWS account? Separate accounts, central logging, no root use, MFA, Identity Center for humans, guardrails, encryption by default, public access blocked and findings reviewed.
  • How do you protect data at rest and in transit? Encrypt with KMS keys whose policies you control, enforce TLS, restrict access with roles and bucket policies, and log access. Mention key rotation and who can use the key.
  • What do you say when asked "is this PCI or GDPR compliant?" That AWS provides services and reports that can support your controls, and that compliance is the bank's responsibility and judgement, confirmed by its compliance and legal teams. Never say "this is compliant".

Cost

  • What causes surprise AWS bills? NAT gateway data processing, idle load balancers and public IPv4 addresses, log ingestion, loops, cross-AZ transfer and forgotten resources. See the bill spike playbook.
  • How do you control cost in a team? Tags in IaC, budgets and anomaly monitors per team, a regular review of top cost movers, and design reviews that include cost.

Troubleshooting scenarios to rehearse

Say each answer aloud in the scenario shape.

Scenario promptWhere the method leadsPlaybook
"A customer was paid twice."Idempotency key, retries, conditional writes, reversalDuplicate payment on retry
"The nightly batch is late."Slow stage, concurrency, skew, safe buy-timeEnd-of-day batch overrun
"The partner API keeps timing out."DNS, connect, TLS, first byte, NAT, retriesPartner API timeouts
"Who looked at this customer's data?"Authority, evidence layers, join, gapsWho accessed this customer record?
"The bill doubled."Account, service, usage type, resource, causeUnexpected bill spike

AWS Console instructions

Not applicable: this lesson has no console steps. If a hands-on screen is part of your interview, be able to navigate to these without a guide: CloudTrail β†’ Event history, CloudWatch β†’ Logs Insights, VPC β†’ Flow logs and route tables, IAM β†’ Policy simulator, Billing and Cost Management β†’ Cost Explorer.

AWS CLI

Be able to type, or at least explain, these. A hands-on round often asks for one or two.

# Who am I, and in which account? The first command in any investigation.
aws sts get-caller-identity --profile learn-sandbox

# What does this role's trust policy allow?
aws iam get-role --role-name hb-payments-task --profile learn-sandbox

# Simulate whether a principal may perform an action (replace the ARNs).
aws iam simulate-principal-policy \
  --policy-source-arn arn:aws:iam::111122223333:role/hb-payments-task \
  --action-names s3:GetObject \
  --resource-arns arn:aws:s3:::hb-statements-111122223333/statements/cust-4471/2026-10.pdf \
  --profile learn-sandbox

# Why did an ECS task stop?
aws ecs describe-tasks --cluster hb-payments --tasks <task-id> \
  --query 'tasks[].[stoppedReason,containers[].reason]' --profile learn-sandbox

# What did this principal change recently? (management events)
aws cloudtrail lookup-events \
  --lookup-attributes AttributeKey=Username,AttributeValue=alex.example \
  --max-results 20 --profile learn-sandbox

# Last hour of errors in a log group, grouped by message (Logs Insights).
aws logs start-query --log-group-name /hb/payments-api \
  --start-time $(date -d '1 hour ago' +%s) --end-time $(date +%s) \
  --query-string 'fields @message | filter @message like /ERROR/ | stats count() by @message | sort count desc | limit 10' \
  --profile learn-sandbox

Infrastructure as Code

Not applicable: there is nothing to deploy for this lesson. If an interviewer asks you to explain IaC, use the example you built in Module 10 and describe what you would change in review.

Complete code

There is no program here. The useful artefact is a story bank you keep in your notes or portfolio repository. One block per story; keep each under 150 words.

story-bank.md
## Story: Duplicate payment on retry (lab reproduction)

- **Tags:** incident, idempotency, payments, ownership
- **Situation:** A client retried a payment request after a timeout, and two payments were created (reproduced locally in Lab 17.1).
- **Constraint:** Retries are legitimate; we could not just disable them. Money movement needed an audit trail.
- **Action (what I did):** I traced the request IDs, added an idempotency key stored with a conditional write, and wrote a test that fires the same request ten times.
- **Result:** Ten identical requests produced one payment and ten identical responses. The test now runs in the pipeline.
- **Lesson:** Decide the duplicate-handling contract at design time and test it with retries, not only with happy paths.
- **Questions this answers:** "Tell me about an incident", "a time you improved reliability", "how do you design for failure".
- **Honest boundary:** This was a lab, not production. In production I would add a reversal procedure agreed with operations.

Code explained line by line

  1. Tags let you find the right story quickly when the question is "a time you disagreed with a colleague" and you scan for the tag.
  2. Situation and constraint make the story credible: a story without a constraint sounds easy and therefore forgettable.
  3. Action (what I did) forces first-person ownership. "We" hides what you actually did.
  4. Result should contain a fact or number: ten requests, one payment; an alarm that now fires; recovery time. If you have no number, say what changed.
  5. Lesson shows you learn. Interviewers value it as much as the result.
  6. Honest boundary protects you. It states in advance what was practice and what was production, so a follow-up question does not catch you out.

Security considerations

⚠ Learning shortcut β€” not for production

The practice stories here use fictional Harbourline data and lab reproductions. In a real interview, never reveal confidential details of your current or former employer: customer names, internal architecture diagrams, incident reports with real names, credentials or regulatory correspondence. Describe the pattern ("a payment service on a managed queue") and keep specifics generic. Your contract and your reputation are at stake.

  • Confidentiality. If a story depends on confidential detail, change it so the lesson survives without the detail, or choose another story.
  • Honesty about experience. Do not claim certifications, production experience or incident ownership you do not have. Background checks and technical follow-ups find the gap, and the integrity concern outlasts the skills one.
  • Take-home tasks. Never put real credentials, personal data or an employer's code in a public repository to show your work. Use fictional data.
  • Live hands-on rounds. If you are given access to a real AWS account, treat it as someone else's money: do not create expensive resources, do not paste real credentials into chats and delete what you create.
  • Never claim compliance. When asked about PCI or GDPR, speak about controls and responsibilities, not certification of a design.

Banking use case

A fictional candidate, Sam, is interviewing for a cloud engineer role at a retail bank. In the scenario round the interviewer says: "Payments are failing intermittently with timeouts to the sanctions provider." Sam does not know the bank's network. Sam answers in the scenario shape:

"First I would scope it: is it every payment or some, one Availability Zone or all, since when? And I would check the safe mode: when the screening call times out, are payments held or released? If anything fails open I would stop that first, because a payment released without screening is worse than a delayed one. Then evidence: I would time DNS, connect, TLS and first byte from the same subnet as the failing tasks, and look at NAT gateway metrics and Flow Logs for rejects. My ranking: an allow-list problem if only one AZ's source IP is failing, port exhaustion if resets cluster at peak, partner slowness if time to first byte is the long phase. I would test the cheapest first. For prevention I would check timeouts and retry layers, and add a circuit breaker with a hold-for-review path. And I would record the timeline for the incident review."

The interviewer follows up on port exhaustion. Sam says honestly: "I have reproduced that in a lab but not in production. The metric is ErrorPortAllocation, and the usual fixes are connection reuse and more Elastic IPs on the gateway." That answer shows method, honesty and specific knowledge. It scores better than a confident guess.

Common mistakes

  • Reciting service names. Why: nerves and wanting to show breadth. Avoid: answer the question that was asked, with one example.
  • No method in scenarios. Why: jumping to a favourite cause. Avoid: say scope, evidence, hypotheses, test, fix, prevent every time.
  • Saying "we" for everything. Why: team culture and modesty. Avoid: for each story say what you decided or did.
  • Guessing instead of saying you do not know. Why: fear of looking weak. Avoid: "I have not run that in production; here is how I would check."
  • Overlong answers. Why: wanting to show everything. Avoid: stop at 60 seconds and offer to go deeper.
  • Badmouthing a past employer or tool. Why: frustration. Avoid: describe constraints neutrally and focus on what you learned.
  • Not asking questions. Why: thinking the interview is one-way. Avoid: prepare four questions about incidents, change control, on-call and learning.

Troubleshooting

Symptom (in the interview)Likely causeFix
You freeze on a scenarioNo default structureStart with "Let me scope this first" and walk the seven steps aloud
You rambleNo endpointGive the answer in one sentence, then add one reason, then stop
A follow-up goes beyond your knowledgeNormalState the boundary, give the principle you would use and how you would verify
The interviewer seems unimpressed by your storiesStories lack results or your roleAdd a fact (a number, a changed process) and a first-person action
You mix up servicesRote memoryAnchor on what a service is for and the Azure equivalent you know
A hands-on round runs out of timeDoing everythingNarrate your priorities, take the most useful diagnostic first, explain what you would do with more time
You forget a commandMemorySay what it does, name the service, write the closest syntax and mention you would check --help
Salary or level question catches youUnpreparedResearch current market data from several sources and decide your range beforehand ⚠ VERIFY

Cost and cleanup

Cost: free. Nothing here needs AWS resources. If you practise the CLI commands in the sandbox, use only read-only commands and, if you create anything for a hands-on rehearsal, delete it the same day.

# Check that nothing from interview practice is left running in the sandbox.
aws resourcegroupstaggingapi get-resources --tag-filters Key=Project,Values=interview-practice \
  --profile learn-sandbox

Exercises

  • 🟒 Guided: Write six story blocks from the template. Time yourself telling the first one aloud and cut it to two minutes.
  • 🟑 Challenge: Pick ten questions from the bank at random, answer each aloud in under a minute, and score yourself: correct, concise, and with a banking example. Redo the weakest three.
  • πŸ”΄ Live problem: The interviewer says: "You join our team and find that the production account has twelve IAM users with active access keys, three of which have administrator access, and nobody knows who owns them." What do you do in your first week? Describe order, risks, communication and what you would not do.

Assignment

Produce a personal interview pack for your portfolio: a 60-second opening, a story bank of six stories, your ten weakest questions with model answers in your own words, and a one-page "how I troubleshoot" summary. Keep it free of confidential employer detail.

Solutions

Knowledge check

Interview questions

These are the meta-questions candidates should be ready for in any loop.

  • Tell me about yourself. A 60-second structure: present role and scope, the experience that matters most here, one thing you built or fixed, and why this role.
  • Why move from .NET development to cloud engineering? Because you already operate services in a regulated environment, and the platform decisions (identity, networking, delivery, cost) now decide how safely the code runs. Give one example of platform work you already did.
  • Tell me about an incident you handled. Use a story block: situation, constraint, your actions, result, lesson. Include how you communicated and what you changed afterwards.
  • What do you do when you do not know something? Say what you know, state the gap, explain how you would find the answer (documentation, a sandbox test, a colleague), and say you would record it.
  • What would you want to learn in your first six months? Their incident process, their access model, their delivery pipeline and their cost visibility, each with a reason.

Certification practice

Completion checklist

  • I can state the three answer shapes and use each on a sample question
  • I wrote a 60-second opening and timed it
  • I have six story blocks, each with my own action and a result
  • I can run through a troubleshooting scenario aloud with the seven steps
  • I prepared four questions for the interviewer
  • I completed one mock interview and chose one thing to improve

Key takeaways

  • Interviewers score method, honesty and judgement as much as knowledge.
  • Use three answer shapes: definition-and-why, scenario, story.
  • Translate your .NET and Azure experience, and be exact about what was lab and what was production.
  • Say what you do not know and how you would find out.
  • Keep confidential detail out and never claim compliance or experience you lack.

Further reading

Independent learning resource β€” not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.

On this page