How we started
In early 2021, two machine learning engineers left their corporate roles in Belfast to start something smaller and more focused. They had spent years building AI features inside large organisations, watching months of work stall in committee reviews. The idea behind Pinnacle Ai Focus was simple: a compact team that can move from problem definition to a running model in weeks, not quarters.
The first client was a dairy cooperative in County Antrim that needed to predict milk yield per herd. We delivered a gradient-boosted regression model in six weeks. It reduced feed waste by 11% in the first season. That project paid for our second hire.
By 2023 we had grown to nine people and completed projects in logistics route optimisation, insurance fraud detection, and clinical trial document classification. Every one of those systems is still running in production today.
Our mission and values
We exist to make Artificial Intelligence accessible to mid-sized organisations that cannot afford a 50-person data science department. Three principles guide how we work:
Transparency over complexity
We explain every model choice in plain language. If a client cannot understand why the system recommends action X, we have not finished the job. We publish model cards for every deployment that document training data, known limitations, and fairness metrics.
Fixed scope, fixed price
Open-ended AI research burns budgets. We scope work into phases of four to ten weeks, each with a defined deliverable and a fixed cost. If we underestimate, we absorb the difference. If the scope grows, we write a new phase agreement before any extra work begins.
Own the outcome
A model sitting in a Jupyter notebook is not a product. We deploy to the client's cloud account, set up monitoring, write runbooks, and train the internal team to maintain the system. When we leave, the client owns the code, the data pipeline, and the documentation.
Meet the team
We are a deliberately small group. Keeping the headcount tight means every person on a project has direct context and can make decisions without waiting for approval chains. Here are some of the people you will work with.
Ciaran Doherty
Co-founder and ML engineer. Previously at Kainos, where he built fraud detection systems for a UK bank. Specialises in tabular data and time-series forecasting.
Roisin McAuley
Co-founder and NLP lead. Holds a PhD in computational linguistics from Queen's University Belfast. Runs all language-model fine-tuning and evaluation.
Eoin Walsh
Data engineer. Designs the ETL pipelines, manages cloud infrastructure on AWS, and makes sure models have clean, versioned data to train on.
Maeve Brennan
Project manager. Keeps timelines honest and clients informed. Before joining us, she managed software delivery at a Belfast health-tech startup for five years.