Why they come last, and what they are for
The whole bootcamp was built backwards from one interview moment — "walk me through something you built" — and the capstones are the answer to it. They are not more curriculum; they are the composition of everything the curriculum already gave you. Each is a single public repository: the infrastructure as code, the pipeline that ships it, and a README that walks a stranger from the outside of the system to the inside and defends every layer between. Certificates claim; repositories prove. These are the repositories.
Two things separate a capstone from a bigger Build. First, it is measured against a published standard — each Plate is a deliberately scoped version of an architecture Microsoft maintains and names, so you are building the industry's shape at a size one person can finish and defend. Second, it is graded the way architects actually grade: against the five pillars of the Well-Architected Framework — reliability, security, cost optimisation, operational excellence, performance efficiency — rather than a binary "does it run." Score evenly across all five and the work reads as designed; spike on one and gap the rest, and it reads as luck.
Plate I first — it is the workload the other two build on. Plate II proves you can design the estate a workload lands in. Plate III proves you can deploy AI infrastructure, not merely discuss it. On every Plate, draw the line between your scoped version and the enterprise original in the README, before an interviewer draws it for you — naming your own gaps is the senior move the whole exercise is designed to teach.
The three plates
The Enterprise Web Platform
A three-tier application delivered entirely through pipelines — infrastructure as code, federated authentication, an approval gate before production, zero-downtime slot swaps, secrets in a vault, telemetry flowing, and a budget watching the bill.
The Miniature Landing Zone
Governance before workload: management groups, policies that deny rather than warn, hub-and-spoke networking with private resolution, centralised logging, and an access model written down as code rather than folklore.
The AI Workload
An internal assistant on Azure OpenAI, reachable only through a private endpoint, rate-limited at the gateway, its token spend charted on a dashboard, its infrastructure shipped by the same pipeline as everything else.
How the plates stand on the builds
Nothing here is fresh learning; if a component takes days, you are learning something in the capstone that belonged in its class — go back to the class, then return. The plates are composition, and each one draws on the milestones you already shipped.
- Plate Icomposes
- Build III's GitOps pipeline and Build IV's observability, wrapped around the private-application pattern of Build II — now assembled into one workload a stranger can inspect end to end.
- Plate IIcomposes
- Build I's governed subscription, grown into a platform: the hierarchy, the deny policies, and the identity model, scaled from one subscription to an estate a dozen applications could land in.
- Plate IIIcomposes
- Everything — Plate I's web-app baseline, the pipeline, the watch — plus the AI infrastructure of Classes 33–35: a private model endpoint, a metering gateway, and a token dashboard.
Certificates claim. Repositories prove.
When you finish all three, Class Thirty-Seven is where they become the résumé: three lines with receipts attached, each one answering a requirement the job posting circled in red. That is the arc the free Class One promised — two servers in a closet, in the opening story, become a governed, observed, AI-serving platform you can defend under questioning. The capstones are where you prove the promise was kept.
The company was fictional; the engineer is not
Campux Retail carried the whole course — the November peak, the nine-hour outage, the 3× invoice, the storefront that grew from a closet to a platform. It was always a teaching device. The plates are where you stop building Campux's estate and start building yours: the same patterns, the same reference architectures, but in a repository with your name on the commits and a README a hiring manager will actually open. The point, from the first free page, was never that you would finish a course. It was that you would have something to put on the table when the interview reaches the sentence that used to end it — and now you do, three times over.