About our Artificial Intelligence practice

We started in 2021 with a simple idea: AI adoption should not require a six-figure budget or a PhD on staff. Four years later, that idea still drives every project we take on.

How we got here

Ai Integrity Alliance began when our founder, Rhona Caldwell, left a data-science role at a large Edinburgh financial firm. She had spent five years building machine-learning models that saved her employer millions, but she kept meeting owners of smaller businesses who assumed AI was out of reach for them. A logistics company in Dundee. A craft brewery in Inverness. A chain of dental practices across the central belt. All of them were drowning in manual data work that a well-scoped model could handle in seconds.

Rhona recruited two engineers she trusted and set up shop in Long Trompthorpe. The first project was an invoice-reading system for a wholesale florist. It took eight weeks to build and paid for itself in under four months. Word spread quickly.

By mid-2023 the team had grown to nine. We moved into a converted mill building at 566 Sporer Mount, which gives us enough space for a small server room, a meeting area, and the whiteboards we go through at an alarming rate.

Ai Integrity Alliance office in a converted Scottish mill building

Mission and values

Our mission is to make responsible AI accessible to businesses that do not have dedicated data teams. "Responsible" is the key word. Every model we deploy comes with documentation that explains, in plain English, what the model does, what data it uses, and where its predictions are likely to be wrong. We think that kind of transparency should be standard, not a premium add-on.

Transparency first

We publish model cards for every system we deliver. These cards describe the training data, known biases, accuracy metrics on held-out test sets, and recommended human-review thresholds. If a model is not confident enough to act alone, it flags the decision for a person.

Fixed pricing

Scope creep is the enemy of trust. After the initial audit we quote a fixed price for the agreed deliverables. If the scope changes, we re-quote before doing additional work. No surprise invoices.

Data stays in the UK

All training and inference happen on servers physically located in the United Kingdom. We do not send client data to overseas cloud regions, and we contractually commit to that in every engagement letter.

Measurable outcomes

Before we write a line of code, we agree on the metric that defines success: hours saved per week, error rate reduction, cost per processed document. If the metric does not move, we have not done our job.

The people behind the work

We are a team of nine, split between engineering, project management, and client advisory. Here are the people you are most likely to speak with.

Rhona Caldwell, founder and lead data scientist

Rhona Caldwell

Founder and lead data scientist

Rhona spent five years in financial-services ML before starting the company. She leads model design and handles the technical side of client proposals.

Callum Bryce, engineering lead

Callum Bryce

Engineering lead

Callum manages the build phase of every project. He came from a DevOps background at a health-tech startup and is obsessive about deployment reliability.

Amina Osei, project coordinator

Amina Osei

Project coordinator

Amina is the person who keeps timelines honest. She runs the shared Kanban boards, schedules demos, and makes sure clients always know what is happening next.

Key milestones

A short history of the things that shaped us.

March 2021

Company registered in Scotland. First project signed: an invoice-extraction system for a wholesale florist in Perth.

November 2021

Completed our fifth project. All five clients renewed for ongoing support, giving us the recurring revenue to hire our fourth engineer.

June 2022

Launched our demand-forecasting product after piloting it with a Glasgow food distributor. Their 22% waste reduction became our most-cited case study.

January 2023

Moved into the converted mill at 566 Sporer Mount. Installed an on-premises GPU cluster for training sensitive models without cloud dependency.

September 2023

Team reached nine people. Introduced formal model-card documentation for every deployment, making bias and accuracy reporting a default part of delivery.

April 2024

Passed 30 completed projects. Median client payback period confirmed at seven months across the full portfolio.

2025

Working on three concurrent builds, including our largest engagement to date: a multi-site document workflow for a Scottish NHS trust procurement team.

If you want to discuss how we could help your organisation, call us on 0500 294134 or email [email protected]. We are happy to have an initial conversation at no charge.

Send us a message