1. LandOffer
  2. Blog
  3. All articles
  4. Cloud Engineering Portfolio: Document a Small Deployment and Its Operating Limits

Job search

Cloud Engineering Portfolio: Document a Small Deployment and Its Operating Limits

A completed fictional deployment record and an executed offline consistency audit show how to explain access, operation, cleanup and unresolved cost evidence.

9 min readLandOffer team

A cloud engineering portfolio should make a small deployment reviewable: what runs, who can reach it, what you tested, what it costs and who removes it. Pair the architecture diagram with access-test results, an operating record and a cleanup decision. Keep the scope small enough that you can explain the architecture, access decisions and operating limits while keeping lab work separate from production experience.

This guide checks official AWS documentation on October 21, 2026. Its deployment case is entirely fictional. The only executed artifact is an original AI-authored offline configuration checker; no cloud account, deployment, credentials, billing or permissions were accessed. LandOffer publishes this guide and offers job-search tools; browse recent roles to choose the engineering responsibility your example should demonstrate.

A small cloud architecture model is paired with an access boundary and a scheduled cleanup record

Choose a deployment whose boundaries you can explain

A useful small project answers a specific request. For example: an authenticated reader returns a sanitized status document from object storage. That creates concrete questions about invocation, storage access, failure handling, logs and teardown. It does not require a public dashboard, multiple regions or a cluster added only to fill the diagram.

Document the request path and the management path separately. In the example used here, a permitted caller reaches a Lambda function URL; the function reads one S3 object and emits logs. A deployment owner manages resources and cleanup. The function's runtime role should not quietly inherit the owner's administrative authority.

An authenticated endpoint is also different from a private network endpoint. The function URL configuration guide says these endpoints use the public Internet and do not support PrivateLink. IAM authentication does not make that a private network endpoint. Explain the boundary you actually configured. Avoid labeling every protected resource “private” when you mean that unauthenticated invocation is denied.

For your real portfolio, identify the account type, region, service configuration and checked date. Preserve a readable case note so reviewers do not need your account or an active endpoint. If the cloud resources are already retired, say so and provide the permitted evidence from when they existed. A closed demo can still support a careful engineering discussion.

Show which account and cost conditions apply to your project

AWS account experiences are changing. The current native Free Tier FAQ describes an October 2026 sign-up experience, temporary credits and a Free plan ending after six months or credit exhaustion, whichever comes first. It also describes a Paid-plan project spend-limit feature starting at $20. These are account-route conditions, not proof that your existing account has every new control.

The inspected page contains different credit-expiry wording across general, upgrade and activity sections. Record the expiration actually shown for the relevant credit instead of promising one universal period. This guide does not resolve those differences by inventing an account result.

For a paid account, list chargeable usage categories: function requests and execution, storage and requests, logs, transfer, and any additional service you enabled. Use current regional pricing and measured usage for a real estimate. A promotional balance is neither a permanent price nor proof that every charge is covered.

An ordinary AWS Budgets notification is not the same as the newer project spend-limit feature. Budget data and notifications can lag incurred usage, and charges can exceed the notification threshold. Name the control you actually used; do not call an alert a hard cap. A portfolio should preserve both gross usage charges and applicable credits when reconciling its eventual cost.

Worked deployment record: a fictional status reader

Fictional candidate Casey's project demonstrates authenticated retrieval of a small status document. The following is a completed hypothetical record, including invented tests and dates. It is not a deployment performed or observed for this article, and the amounts are personal planning thresholds rather than AWS price quotes.

Project field Completed fictional record Evidence boundary
Scope and dates Status reader, October 6–8, 2026; us-east-1 One short-lived regional demo, no availability commitment
Account Existing Paid account; no credits assumed New sign-up controls and credit eligibility not inferred
Architecture Function URL → Lambda → one synthetic S3 status object; CloudWatch logs No database, custom domain, customer data or multi-region design
Runtime 128 MB, three-second timeout; seven-day log retention selected Example configuration, not a sizing recommendation
Caller boundary AWS_IAM; caller permission record includes both invocation actions Authenticated URL, not a network-private endpoint
Data boundary Read only the designated object; bucket not public Runtime access distinct from deployment-owner privileges
Operations Casey owns the alert review, failure note and teardown No team or on-call service implied
Cost condition $5 planning budget, $1 ordinary alert; final bill unreconciled Neither amount is a cap or a measured total

The fictional test record contains three checks: an authorized request returned the synthetic status; an unsigned request was denied; a request for a different object was denied. Casey records those as the stipulated outcomes of this hypothetical example. A real applicant needs their own dated evidence, including the tested caller and resource scope, before making equivalent claims.

Casey's completed architecture decision is to retain authenticated invocation and share the case note instead of opening the endpoint for a recruiter. The trade-off is that a reviewer cannot simply click the demo without authorization. The benefit is that reviewability does not depend on widening cloud access. Screenshots or excerpts used in a real case must omit tokens, account identifiers and sensitive request content.

Casey's final project blurb reads: “Short-lived AWS status-reader demonstration in us-east-1, with authenticated invocation, scoped object access and separate log cleanup. The case record documents access checks and retirement; final cost reconciliation is pending.” It does not claim a zero-dollar system, production reliability or a service still running today.

Show the access decision and its unsuccessful case

The current Lambda function URL access guide distinguishes AWS_IAM from NONE authentication and describes policy evaluation. It also notes that new function URLs require both lambda:InvokeFunctionUrl and lambda:InvokeFunction permissions. Merely showing an auth-type field does not demonstrate that the full caller and resource-policy configuration works.

For a real deployment, preserve a sanitized configuration excerpt and the corresponding permitted and denied test cases. Separate the caller's invocation permissions, the function's runtime data permissions and the owner's deployment permissions. Logging permissions are another necessary consideration; a storage-access snippet alone is not a complete execution-role policy.

Explain an expected denial in terms of a request and a boundary: “This caller could invoke this function, but the function could not read the alternate object.” Do not turn a single denial into a security certification. Other identities, cross-account policies, organization controls and configurations can change effective access.

Keep cleanup as part of the engineering record

Deleting the main function does not prove that all related resources are gone. AWS's Python Lambda logging guide explicitly says function deletion does not automatically remove its log group. The CloudWatch retention guide also distinguishes retention settings from immediate removal; logs normally remain indefinitely without a selected policy.

Storage needs its own inspection. In a versioned S3 bucket, a simple delete can insert a delete marker while older versions remain. The object-version deletion guide explains why removing the visible object is insufficient evidence that its stored versions disappeared. Cleanup must match the resources and protection settings you actually used.

In Casey's fictional retirement ledger, October 8 entries record removal of the function URL and function, separate log-group cleanup, review of object versions and delete markers, and retirement of the dedicated bucket, role and inline policy. The dedicated budget-notification entry is reviewed separately so the retired demo does not leave an unexplained alert behind. Those are stipulated example entries, not AWS responses captured by this article.

The ledger ends with two different conclusions: resource retirement is recorded in the fictional case; final billing is still pending. Casey therefore removes the live-demo claim and withholds a total-cost claim. For your own case, preserve actual inventory and billing readbacks after the cleanup rather than assuming a successful delete command settles every resource or delayed charge.

Worked offline check: catch contradictions before claiming a deployment

This article actually ran a local Python checker against eight original synthetic manifests. The custom teaching schema describes Casey's intended region, account plan, access choices, retention, cost language and cleanup inventory. It is not Terraform, CloudFormation or an AWS policy document, and it cannot provision anything.

The baseline manifest declares AWS_IAM authentication, both invocation action names, object-scoped s3:GetObject, a nonpublic bucket, seven-day retention, no permanent-free claim and separate cleanup entries. Credit balance and expiration are unknown, and final billing remains unverified. Passing means only that these declared fields satisfy this checker's local rules.

The completed mutation audit is:

Local case Checker result Corrective decision
Baseline Matches declared rules Keep as a draft specification, not cloud proof
Change auth to NONE Unexpected public-auth choice Restore intended auth before any real deployment decision
Mark bucket public Data exposure contradicts scope Restore nonpublic-data intent
Replace object resource with wildcard Data scope expanded Retain the specified object boundary
Omit log group from cleanup Cleanup incomplete Add a separate log-group entry
Call the alert a hard cap Cost-control description false Label it as a notification
Omit one invocation action Required action list incomplete Reconcile both documented action names
Claim permanently free Cost claim unsupported Remove the perpetual-cost promise

The executed run accepted one baseline and flagged seven changed cases, with one expected finding in each. The saved script passed ten assertions, including an unchanged input-file SHA-256 hash and a check that final billing remained unverified. The saved report records Python 3.12.14. It used no AWS SDK, credentials, network calls or listening server.

This check can catch a contradiction in a case note before the candidate repeats it in an application. It cannot evaluate IAM's complete authorization logic, service availability, quota behavior, actual resource deletion or regional pricing. A local pass does not convert the hypothetical deployment into a real cloud achievement.

An offline project specification is checked against access, cleanup and cost-boundary evidence without creating cloud resources

Publish a readable case, then describe your actual contribution

A reviewer should be able to follow the request path, understand one access denial, inspect a failure or limitation, and find the retirement and cost status. Include your role and artifact version. If someone else wrote the infrastructure configuration or operated the deployment, identify that contribution rather than presenting the entire result as personal implementation.

Retain a fallback containing the architecture, sanitized excerpts, finite checks and operating decisions. Do not require a recruiter to create a cloud account or receive elevated access to understand your work. A screen recording, when permitted and free of secrets, can supplement the case; it should not replace the written configuration and result boundaries.

As you compare recent roles on LandOffer, match the opening’s access-design, deployment, debugging, operations and cleanup responsibilities to evidence you can actually show. Describe a lab as a lab, a fictional example as fictional and a deployed system only when you have actual deployment evidence. Use the operating record to give reviewers concrete evidence of the decisions you made and the limits you retained.

Sources and Further Reading