Applied Infrastructure · Live Client Systems
We Design Deploy and Maintain
Governance Framework.
Two real infrastructure builds, explained plainly — what they do, how they're put together, and exactly where their boundaries sit. No black boxes, no buzzwords.
01Case Study — Golden Drop Mango Winery
Bella, the Biboohra Liquid Chef —
Anatomy of a Stateless Agent
goldendrop.com.au · live production system

We don't claim to build sentient AI or flawless autonomous systems. What we build are tightly bounded, stateless micro-agents designed to solve one distinct, practical problem well. Bella is one of them — an interactive cocktail-mixing feature built into Golden Drop Mango Winery's online shop, built to demonstrate the discipline behind it as much as the feature itself.

A customer selects two or three bottles they already have, or fancy trying. Bella returns two real serving suggestions, tailored to those exact bottles, and puts the ingredients straight into the cart. That's the whole interaction. Here's what's actually running underneath it.

It's easy to say "automation pipeline" and move on — but that hides more than it explains. So here's the actual chain of machinery behind Bella, start to finish, in plain terms. Because Bella's a chef, the easiest way to picture it is as a real kitchen.

01
The Kitchen Itself
Amazon Web Services · EC2 Instance
This cloud-based computer isn't shared with anyone else's website — it's the Golden Drop Kitchen, and only Golden Drop's orders ever run through it. It's built with proper, commercial-grade memory and processing power behind it — the same class of machine a real business runs on, not a bargain-bin server — so it can comfortably handle a couple of customers on a quiet afternoon, or a whole tour bus of them all asking Bella for a drink at once, without slowing down.
02
The Front Door
Nginx
Think of it like walking into a busy restaurant — there's a host at the door who seats people as tables become free. If a few customers arrive at once, there might be the smallest pause while everyone's shown to a seat in turn. That's what's happening here: incoming requests are managed and evenly allocated, so the kitchen behind it is never asked to do more at once than it can handle properly.
03
The Kitchen Line
n8n Automation Engine
Once inside, the order has to move from station to station — "customer picked these bottles" → "recipe drafted" → "ingredients added to cart" — without a person physically carrying the ticket between each step. This is that line: it passes the ticket along instantly and correctly, every single time.
04
The Chef, Working to House Rules
Google Gemini · Stateless AI Instance
At the pass stands the chef — Bella herself, built on a Gemini model — but she isn't working from her own imagination. She's handed a strict house recipe book containing only Golden Drop's own bottles and common household mixers. She doesn't get to wander off and invent something that isn't actually on the shelf.
05
The Till
Shopify API Integration
The finished plate still has to become a real sale. This last connection takes Bella's recipe and turns it directly into real bottles, sitting in a real cart, on the same checkout every other Golden Drop purchase runs through.

We call this a "pipeline" because it's the closest ordinary word for it — even though, like most engineering shorthand, it doesn't quite capture the whole picture. Now you've seen what's actually inside it.

Why Stateless?

A balanced technical trade-off, not a marketing claim.

The Limitation
Bella doesn't hold an open-ended, organic conversation. Each interaction is its own clean calculation — she isn't designed to remember what was said earlier in the same conversation.
The Benefit
By deliberately not carrying memory between requests, the surface area for drift and hallucination is drastically reduced. Every recipe is calculated fresh, strictly from the current selection and the business's own constraints — predictability over conversational flair.
Human-in-the-Loop
Software Is Never "Set and Forget"
Shopify's own systems change over time, real customers present edge cases no one anticipated, and language-model behaviour needs ongoing observation, not a one-off launch and a walk away. We actively monitor Bella's pipeline, refine her operating boundaries as needed, and keep the infrastructure stable and aligned with how Golden Drop actually operates — the same standard of oversight we apply to every system we build, including this one.

02Case Study — Independent FNQ Retailer
Automated Invoice Ingestion for
a Legacy Retail Operation
motorcycle & apparel retail · Far North Queensland
Ingestion — Live & Tested Connectivity — In Progress With Client

A well-established Far North Queensland retail business — running on a legacy stock and invoicing system — needed a way to bring incoming supplier paperwork into their digital systems without re-keying every line by hand. This build uses the same underlying pipeline architecture as Bella — bounded, monitored, stateless where it needs to be — applied to a completely different problem.

What's Live Now
Ingestion
A staff member photographs the invoice or packing slip accompanying an incoming stock delivery on their phone. The pipeline reads it, extracts the relevant line items, and prepares them for the next step — tested and running today.
What's In Progress
Onward Connectivity
We're working directly with the client to finalise where that captured data actually needs to land — including their accounting platform for creditor recognition and their e-commerce stock system — so the handoff matches how they actually run the business, not a generic template.
Human-in-the-Loop
Built With the Client, Not Just For Them
This is a live example of what "applied" actually means for us — a real business, a real legacy system, and connectivity being shaped in direct collaboration with the people who'll use it every day, rather than a fixed solution handed down and hoped for.
This is what "applied" means to us.
Curious what it could mean for your business?