A small, focused team in Scotland that has spent the last seven years turning messy data into production-grade AI systems.
Mastery AI Quest started in 2018 as a two-person experiment. Rhona Caldwell, a machine-learning researcher at the University of Edinburgh, and David Osei, a software engineer who had spent a decade building backend systems for fintech startups, kept running into the same problem: companies wanted to use AI but had no clear path from a proof of concept to a live, monitored system.
So they built one together for a mid-size haulage firm in Glasgow. The project was a demand-forecasting model that predicted next-week delivery volumes with 91% accuracy. It went live in eight weeks. That haulage firm is still a client today.
Since then the team has grown to twelve. We have delivered 87 projects across logistics, financial services, retail, manufacturing, healthcare administration and several other sectors. Every project follows the same discipline: fixed scope, fixed price, weekly check-ins, and a working system at the end.
We are not trying to be the biggest AI consultancy in the UK. We want to be the one that actually ships.
These four principles shape every engagement, from the discovery call to the 90-day support window.
If a rule-based script solves your problem, we will tell you that on the discovery call and save you the cost of an ML project. We only recommend AI when the data and the business case justify it. About one in five enquiries ends with us suggesting a simpler approach, and we consider that a good outcome.
Open-ended research engagements drain budgets. We agree a delivery date before work starts, and we hit it. Over 87 projects our average overrun has been 2.3 days. When delays happen, they come from data access issues on the client side, and we flag those in writing within the first week.
At the end of every project you receive the full source code, model weights, training scripts and documentation. No vendor lock-in. If you want to bring model maintenance in-house after the support window, we run a two-day handover workshop with your engineering team at no extra charge.
Every model we deploy comes with a bias audit report and an explainability layer. For classification systems we document precision, recall and fairness metrics broken down by relevant demographic groups. We follow the UK ICO guidance on AI and data protection, and we keep records of every training dataset used.
Twelve people, four time zones, one Slack workspace. Here are the leads.
Co-founder and ML lead
Former research associate at Edinburgh Informatics. Rhona designs model architectures and runs the data-science team. She has published twelve peer-reviewed papers on time-series forecasting and spent two years at a healthcare analytics startup before co-founding Mastery AI Quest.
Co-founder and engineering lead
David writes the infrastructure code that keeps models running in production. Before Mastery AI Quest he built payment-processing pipelines handling over two million transactions a day. He cares about uptime, monitoring and deployment automation.
NLP specialist
Priya leads our natural-language projects. She fine-tunes transformer models for document classification, entity extraction and conversational agents. Her background includes three years at a legal-tech company where she built contract-analysis tools used by over 200 law firms.
Computer vision lead
Marcus handles everything from image annotation strategy to edge deployment on NVIDIA Jetson devices. He spent six years in automotive R&D working on object-detection systems for driver-assistance features before joining the team in 2021.
A short history of the company told through the moments that shaped us.
Rhona and David deliver a demand-forecasting model for a Glasgow haulage company. The model goes live in eight weeks and is still running today with periodic retraining.
Three new hires join: a data engineer, an NLP researcher and a front-end developer to build client-facing dashboards. Revenue crosses £300k for the first time.
During the pandemic we build a patient-flow prediction tool for an NHS trust in the Borders region. The model helps administrators anticipate bed occupancy 72 hours ahead with 88% accuracy.
Marcus Beattie joins and we add computer vision as a formal service line. First project: defect detection on a bottling line for a Scottish distillery.
We pass 80 completed engagements. Client retention rate sits at 97%, and the team has grown to twelve people across Scotland, London and Lisbon.