Restaurant Automation Platform

Four Applications, Four Different Users, One API

Section 1 of 4

A restaurant is not one piece of software. The diner choosing where to eat, the waiter taking an order at a table, the kitchen working the ticket queue, the owner reconciling stock against last night's sales, and the platform operator onboarding a new venue are five different jobs with almost nothing in common except the order flowing between them.

This project is the attempt to build all of it: a multi-tenant restaurant operations platform with four distinct front ends over a shared Node and MongoDB API. A customer progressive web app for discovery, reservation and ordering. A waiter and kitchen terminal for table service. A restaurant back-office covering inventory, purchasing, costing and profit and loss. And a platform operator console for onboarding venues, managing vendors and running a support desk.

It is also the oldest substantial project in this portfolio, and the sections below treat that honestly. The breadth is real and so are the shortcomings — the authorization model in particular is one I would build very differently today, and the Retrospective section says exactly how and why rather than leaving it to be discovered. That contrast against the later work here is the most useful thing this case study has to offer.

4

front-end applications

~41K

lines of application code written

9

domain routers behind one API

67

custom back-office screens

1

shared MongoDB data layer

System Architecture

Four applications serving four roles, every one of them talking to the same Express API over MongoDB. The surfaces differ enormously in what they show; underneath, they are reading and writing the same orders.

Customer PWAdiscover, reserve,order, payWaiter & Kitchentable service,ticket queueBack-Officestock, purchasing,costing, P&LOperator Consolevenues, vendors,support deskExpress APIone service, mounted under a single versioned route prefixNine Domain Routersauth · users · customers · menu · categoriesbanners · coupons · orders · QR codesMongoDB via Mongoosetenant-scoped documents · Stripe for payment intentsEvery surface reads and writes the same order documents — the difference between them is which slice of the lifecycle they exposeThe two operations consoles are built on a commercial React admin template; the domain screens inside them are not

What the Platform Actually Covers

A single order is created by a diner on their phone or a waiter at a table, appears in the kitchen queue, gets marked ready, is billed and paid, lands in the night's sales figures, decrements ingredient stock, and shows up in the operator's platform-wide revenue reporting.

That is one record crossing four applications and four very different user intentions. Designing it as four separate products with a shared database — rather than one product with four skins — is what lets the kitchen screen show a stripped-down ticket queue while the back-office shows the same order inside a profit and loss calculation.

A note on scope and source

The figures above count application code I wrote: the API, the operator console, the customer PWA and the custom domain screens inside the two template-based consoles. They deliberately exclude the admin template's own component library, which is licensed third-party code and would multiply the total several times over without representing work done.

This case study describes architecture and design decisions rather than linking to source. It is an older project carrying configuration that predates how I handle credentials now, and the Retrospective section covers that directly.