I work across the full data lifecycle — business intelligence, data engineering, and data science. Most of my work lives in Power BI, where I design and optimise semantic models built to stay fast and cost-efficient at scale. Alongside that I build and maintain the pipelines that feed them, and I reach for Python, machine learning, or LLMs when they solve a problem better than a dashboard can.
What I enjoy most is the messy end: taking unreliable data and a vague business question, and turning them into something people actually trust and use.
One of the world's market leaders in drainage technology.
One of the world's largest development and construction companies.
Specialist high-risk motor insurance underwriter.
Four areas, and I work across all of them — which is the point. The reporting layer, the pipelines underneath it, the models on top, and the people and budgets that keep it running.
An agentic AI app that turns Reddit discussions into clear, research-style insights. It pairs web search tools with a lightweight GPT-4o-mini model to analyse subreddit activity, surface themes, summarise sentiment and highlight emerging topics — a demonstration of how tool use makes even a small model genuinely useful for practical analysis.
Four CNN models built to classify brain MRI scans for Alzheimer's disease and brain tumours, by severity and tumour type. The strongest results came from ResNet50 and EfficientNetV2S with data augmented via Albumentations. I combined them into a Streamlit app that classifies an uploaded scan and returns the predicted class and probability.
A team forecasting project combining Google Trends with state-mandated restriction data. We modelled search behaviour with SARIMAX — with and without restrictions as exogenous features — and used SARIMA and Greykite to test how the onset of restrictions changed mental health related searches.
An NLP model that distinguishes Bioinformatics from Data Science Reddit posts, collected via PushShift. A stacking model combining Naive Bayes, XGBoost and Gradient Boosting reached an F1 of 0.85 and 91% test accuracy — a large gain on the 67% baseline.
I teach Power BI performance optimisation. DragoFab is my channel for practical tips, tricks and tutorials — interactive dashboards, paginated reports, DAX techniques, automation, and storytelling with data. It's where I work out how to explain the things I've just spent a week fixing.
Based in Milton Keynes, UK — with the right to work in the UK, no sponsorship required. Open to conversations about senior data roles across BI, engineering and analytics leadership.