Hi, I'm
Data Scientist & Machine Learning Engineer
MSc Data Science graduate (Distinction, University of Essex) specialising in predictive modelling and multimodal deep learning for health and sensor-driven domains. I design end-to-end ML pipelines — from signal preprocessing and feature engineering through model benchmarking to REST API deployment — with a track record of beating baselines across time-series forecasting, biosignal classification, and computer vision. Proficient in Python, TensorFlow, PyTorch, and Scikit-learn, based in Colchester, UK.
A selection of applied data science and machine learning work, spanning healthcare, sensor data, NLP, and business analytics.
Multimodal framework forecasting post-meal glucose spikes at 1, 3, and 6-hour horizons by fusing CGM time-series with dietary and biomarker data. Raised 1-hour prediction R² from 0.58 to 0.68, a 17% improvement over the gradient-boosting baseline.
Multi-class mental workload detection from raw EEG biosignals, with a full preprocessing pipeline spanning filtering, epoch segmentation, and spectral feature extraction. Deep learning lifted 5-fold CV accuracy from 68.3% (logistic regression) to 93.6%.
UK healthcare operations analytics: NHS England A&E attendance volumes and 4-hour waiting time standard, built with real NHS statistics (SQL + Excel + Power BI).
Flutter mobile app with on-device TensorFlow Lite models delivering real-time agricultural recommendations to farmers — 98.4% accuracy on crop recommendation and 90.1% on CNN-based potato disease classification.
Predicts UK sale prices from 2.7 million real Land Registry sales. Trained on 2023-24 and tested on unseen 2025 sales with leakage-safe target encoding; gradient boosting cut median error from 30.6% (district-median baseline) to 19.9%, with error analysis by region and honest limits (no floor area in the data).
End-to-end pipeline processing 2.77 million HM Land Registry house sales: automated download, cleaning and pandera validation, an idempotent load into a SQL star schema, reporting views, and a Power BI dashboard spec with DAX. Outliers are flagged rather than deleted, and a CI-tested codebase (33 tests) runs on PostgreSQL, SQLite or SQL Server.
Retrieval-augmented generation pipeline delivering grounded, cited Q&A answers over UK government policy documents.
End-to-end churn pipeline on 7K customers — EDA, feature engineering, and a Logistic Regression vs. Random Forest comparison with business retention recommendations.
Interactive Streamlit + Plotly dashboard over a 15K-row e-commerce dataset with live-filterable KPIs, revenue trends, and a top-products leaderboard.
End-to-end analytics portfolio combining time-series forecasting, RFM/K-Means customer segmentation, NLP sentiment analysis, and SHAP-explained churn prediction.
Power BI star-schema report on cost of living and regional inequality, built on real ONS, Land Registry, and MHCLG public datasets.
Full exploratory analysis of a 10K-order retail dataset — univariate, bivariate, and time-series analysis, a correlation heatmap, and 6 data-backed business insights.
Full statistical analysis of a landing page redesign test on 10K visitors — chi-square and z-tests, power/sample-size validation, and a rollout recommendation.
Self-contained SQL analytics project — 15 business queries (CTEs, window functions, cohort analysis) over a synthetic retail database, exported to a formatted Excel report.
Automated pipeline turning a raw sales CSV into a polished multi-page PDF report with KPIs, charts, and week-over-week growth tracking, built to run unattended on a schedule.
Data-cleaning project on a deliberately messy 5K-row HR dataset — 9 documented cleaning steps, fuzzy department matching, and a before/after data-quality report.
Passed with Distinction.
Graduated Cum Laude.
Open to Data Scientist and Machine Learning Engineer roles. Feel free to reach out.