Work

Systems we have built

Real systems, described from their code without client names. Fuller case studies will follow where a client gives written permission.

So far: industrial manufacturing, wholesale distribution, cross-border healthcare, and a consumer product of our own.

Industrial manufacturing

Three connected systems for one manufacturer

The work began as a ticket system for staff and grew into the systems the company runs on. We keep developing the portal and the operations system, both in daily use.

1
In production, in daily use

Ordering portal for trade buyers

What it does
Buyers see their own prices and current stock, place orders and download invoices. Orders go straight into the accounting system, so nobody retypes them.
How it's built
A web app for buyers in three languages, working from a copy of the manufacturer's data that refreshes every few minutes. Each order carries its own reference, so a retry can't create a second order, and prices are always worked out by the server. A full re-check runs daily, because the accounting link can't report deletions.

Buyer

Portalprices worked out on the server

Order with its own reference

Accounting system

Data copy refreshed every few minutes · full re-check daily
Our role
Built from scratch. We still develop it.

Built with
Node.js, Express, PostgreSQL, plain JavaScript
Size
30+ database tables
15+ screens
3 languages
10+ acceptance test suites
Built for the same manufacturer as the operations system below.
2
In production, in daily use

Operations system for staff

What it does
Sales, contracts, purchasing, warehouse and finance in one place, with figures straight from the accounting system. More than 20 job roles each see only what they need.
How it's built
One web app for staff that began as a ticket system for IT requests and now reads the company's ERP database without changing it. It also connects to banks, a delivery service, public tenders, company registers, messengers and the phone system. Access rules for every module sit in one list, and the menus and permission screens are built from it. The database is SQLite in WAL mode on the application server, so there is no separate database server to look after.

ERP databaseread only

Operations systemone access list for every module

Banks · delivery · tenders · registers · messengers · phone
Our role
Built from scratch. We still develop it.

Built with
Node.js, Express, SQLite, plain JavaScript; tests with Jest and supertest
Size
260+ database tables
1,000+ API endpoints
70+ scheduled jobs
500+ automated test files
Built for the same manufacturer as the ordering portal.
3
Delivered

Sales call checks

What it does
Every sales call is transcribed and checked by AI against the same criteria, with a report for the sales manager.
How it's built
When a call ends, the phone system notifies the service, which fetches the recording and transcribes it. The AI returns its scores as separate fields, not prose, so they can be counted and compared over time. Every AI call is logged with its cost, and managers sign in to the dashboard with their Google account.

Phone systemcall ends

Recording

Transcript

AI scoresas separate fields

DashboardGoogle sign-in

Each AI call logged, with its cost
Our role
Built from scratch.

Built with
Python, FastAPI, Docker on Google Cloud Run, OpenAI speech-to-text, Claude

Built for the same manufacturer's sales team.

In production

SectorCross-border healthcare provider

Website in four languages

What it does: Clinics, doctors and treatments managed in one place and published in four languages. Each patient enquiry goes straight to the right clinic.

How it's built

A custom WordPress theme with its own content types for clinics, doctors, treatments, procedures, programmes and articles. The four languages are built into the theme, with no translation plugin. Enquiries are stored privately and sent by messenger and email, and the patient confirms each one through a messenger link.

Our role: Built the site, working alongside the provider's own developers.
Built with: WordPress, PHP, Algolia site search
Handed over: Recent changes went to the provider's developers as code, with written instructions.
Size: 4 languages
6 content types
750+ interface phrases per language

Delivered

SectorEU wholesale distributor

Product website in eight languages

What it does: A catalogue site in eight languages, rebuilt from the distributor's product file whenever the range changes.

How it's built

A static site, so pages load fast and there is little to look after on the server. Each language has its own web addresses, and a visitor who switches language stays on the same page. The quote form posts to a small script on the distributor's own hosting.

Our role: Built from scratch.
Built with: Astro, TypeScript, PHP for the quote form
Handed over: Deployed to the distributor's own hosting.
Size: 8 languages

Our own product and tools

Built and run by us, not for a client. They are here because billing, updates, costs and support are our own problem too.

Our own product

In production, early stage

SectorEndurance sports

Training and coaching platform

What it does: Athletes get a season plan and weekly training sessions that adapt to data from their sports watch, with a matching meal plan.

How it's built

A web app that installs on a phone, with subscriptions billed through Stripe. The AI coach returns plans as checked, structured data, and every AI call is logged with its cost. A release fails if any of the ten languages is missing a required phrase.

Sports watch data
AI coachchecked, structured plans
App on the phoneten languages

Subscriptions billed through Stripe

Built with: Python, FastAPI, SQLAlchemy, React, TypeScript, Stripe, Claude
Size: 35+ database tables
140+ API endpoints
40+ database migrations
10 languages

Our own tools

Bookkeeping import: bank statements read, checked against their own balances and posted to Xero in batches. A final check stops on any mismatch.
Customs data analysis: large import and export datasets converted into compact files for fast analysis.

Sizes: measured from each system's own code, rounded down.

How we describe our work

Each entry gives the sector, what the system does for the people who use it, and how it is built. Client names, logos, web addresses and business figures are left out.

We don't show screenshots, because they would show real customer records.

Our tools and how we keep systems running