Software engineering, data and AI · Headquartered in Scotland, delivering internationally| Start a project →
← All work
Case study · Other

Building AANDA Labs itself

AANDA LABS LIMITED

The challenge

AANDA Labs needed a commercial site that would do three jobs at once. It had to sell the studio's four services in writing. It had to carry enough institutional credibility to open doors with grant committees, investors and professional reviewers — including UK Global Talent Visa Digital Technology endorsers. And it had to be built to the same production-grade standards we would apply to a paying client's project, so the site itself could be a working case study of what we ship.

The usual shortcut for a brand-new consultancy — a templated marketing site on Webflow or WordPress — was ruled out by the second requirement. Reviewers notice. A static templated site says "we have nothing to show yet and are relying on words." A custom production system with its own admin, its own commerce engine, and its own audit trail says the opposite.

What we built

We built the site as a full Django 5.2 application rather than a marketing site, using the same opinionated patterns we apply to client platforms.

The data model is organised across twelve apps: core, accounts, audit, services, products, orders, subscriptions, quotes, portfolio, blog, analytics, notifications. Each follows the Django convention of self-contained models, views, URLs, templates and admin, with shared primitives — BaseModel, AuditableMixin, role hierarchy — living in the core app. Every customer-facing table uses UUID primary keys, created-by/updated-by foreign keys, soft-delete via is_deleted plus deleted_at, and timestamps. No hard deletes anywhere in the system.

Authentication is email-only via django-allauth. The User model carries a role field with a numeric hierarchy (admin 100, director 90, manager 70, staff 30, client 10) and a stripe_customer_id for payment linkage. Role checks go through a shared @role_required decorator and a can_view() helper that consult the numeric hierarchy — higher roles implicitly have the permissions of every lower role.

The audit layer is hash-chained. Every model that inherits AuditableMixin automatically produces an AuditLog row on every save and every delete, via two post_save/post_delete signal handlers. Each AuditLog row's entry_hash is a SHA-256 of a canonical JSON serialisation of the event plus the previous row's entry_hash. Altering any historical row invalidates that row and every subsequent row's hash — tamper-evidence without external infrastructure.

Order and quote numbering follow deterministic human-readable formats: AANDA-YYYY-NNNN for orders, QR-YYYY-NNNN for quotes. Generation happens inside a transaction with a FOR UPDATE lock on the last row for the current year, which is adequate at the write volumes a small studio produces.

Payments go through Stripe Checkout. The orders app includes a /checkout/ view that creates a Stripe Checkout Session server-side and redirects. In development mode with placeholder keys, it instead renders a friendly preview page so the rest of the app is testable without hitting Stripe. A /stripe/webhook/ endpoint verifies incoming signatures and updates Order status on checkout.session.completed and charge.refunded. The whole flow — session creation, webhook, idempotency — is production-ready; only the keys are placeholders until a trading decision is made.

The front end is server-rendered Django templates. There is no React, no jQuery, no SPA. A single base template contains the entire stylesheet in a <style> tag. Every child template uses 100% inline styles on each element — this is a deliberate pattern we have proven across other production Django apps to avoid CSS cascade conflicts with the base layer. Vanilla JavaScript handles the cookie banner and the live-pulse polling pattern we use for authenticated dashboard surfaces.

Settings are split across base.py, dev.py, and prod.py using django-environ. The database is chosen by a DATABASE_URL read through dj-database-url, so the same code runs on SQLite or PostgreSQL; production switches on HSTS, SSL redirects, and Sentry. SendGrid handles production email; Cloudinary handles media. Production runs on an Oracle Cloud server in London behind nginx and gunicorn, with a free Let's Encrypt certificate. Server setup and releases are scripted in the repository, so a rebuild from scratch is two commands.

The commercial surface — the public marketing site that you are reading this on — is a thin UI layer on top of this stack. Every page extends base/base.html, every service pillar maps to a Companies House-registered SIC code (62012, 62020, 63110, 85600), every price is transparent, every legal page is present. The design language is restrained: warm off-white backgrounds, deep neutral text, sparing use of a single sky-blue brand accent, Inter for prose and JetBrains Mono for section labels and numeric data. The intention was to feel comparable to Linear, Stripe or Thoughtbot rather than to shout.

Outcome

The commercial site shipped as a complete production system from first commit. All seventeen public routes return the correct HTTP codes. All migrations apply cleanly against both SQLite (dev) and PostgreSQL (prod). The code is open-sourced in a private GitHub repository ready for clients to review during technical due diligence — they can see exactly how we build before they commission us.

The audit chain pattern is being packaged as a standalone Django library under an MIT licence, which doubles as a public-artefact signal for Global Talent Visa endorsement.

Stripe integration is wired end-to-end but gracefully degraded in dev mode, so demonstrations can be given to prospective customers without a real transaction.

The settings split, the role hierarchy, the BaseModel and the AuditableMixin pattern are the same primitives used in prior client systems. Carrying them forward into the AANDA Labs commercial site means that every new engagement starts from a known base — no re-deciding architecture, no bikeshedding over conventions. New team members will find the patterns consistent with the rest of the firm's work.

Perhaps most importantly: the site is itself the first case study. We can send any prospective client, grant assessor, or investor the link, and they can inspect the repository, read the models, and verify that the production-grade claims on the services pages are backed by production-grade code.

“The AANDA Labs site is the clearest proof-of-work document in the firm's portfolio — it is simultaneously the marketing surface and the engineering demo.”
— The founder, writing honestly

Want something similar?

Start a project →