Seven projects, strongest first

Education Analytics Dashboard

Power BI · Business Intelligence Analysis module · group project

An interactive dashboard analysing student performance, teacher effectiveness and resource allocation across four regional schools — built so a school administrator could see where to spend attention first, rather than reading four separate reports.

65.57% Average attendance across the four schools.

Designed the data model tracking attendance, subject-level failure rates and student–teacher ratios. Ran a correlational analysis of teacher experience against GPA, which showed a positive relationship and exposed performance disparities between regions. Delivered predictive GPA modelling and resource-optimisation recommendations.

• Power BI
• Data modelling
• KPI design
• Correlational analysis

Credit Risk Prediction

Python · scikit-learn · German Credit dataset

A decision tree classifier predicting loan default on a heavily imbalanced real-world dataset. Overall accuracy is the easy number to move; the number that matters is whether you catch the bad-risk applicants you would otherwise approve.

Addressed class imbalance and model overfitting with hyperparameter tuning via grid search cross-validation and pre-pruning. Deliberately prioritised recall on the minority class over headline accuracy — doubling detection of the cases that carry the cost.

• Decision trees
• Grid search CV
• Class imbalance
• scikit-learn

Microsoft × EDI Business Challenge

AI for social good · competition

Two AI-driven products built against real social and governmental problems: Altruize, a donor–charity matching system designed to raise engagement and optimise fundraising, and consultUK, a government-backed generative AI platform for public consultation that compresses policy response time from months to immediate.

Presented to leaders from Microsoft, Mind, the National Trust, Oxfam and RNIB — and pitched consultUK at 10 Downing Street.

The analytical work was in the matching logic and in framing what "success" meant for each product before building anything: engagement and fundraising yield for Altruize, response latency and coverage for consultUK.

• Generative AI
• Product framing
• Stakeholder pitching

Traffic Management & Heart Disease Risk

Python · CN5021 Programming for Data Science · December 2024

Two pieces of work applying the full data science lifecycle. The first: an intelligent traffic management framework using computer vision models — YOLO and U-Net — for vehicle detection and classification, with a deployment strategy for real-world use. The second: a K-nearest-neighbours classifier predicting heart disease risk across a 303-record clinical dataset.

On the clinical model, performance came from hyperparameter tuning and feature scaling rather than from a more exotic algorithm — a small dataset punishes complexity, and the discipline was in recognising that.

• YOLO
• U-Net
• K-nearest neighbours
• Feature scaling

Cryptocurrency Price Prediction from On-Chain Data

BSc final project · Python · LSTM multivariate time-series

A deep learning framework forecasting short-term price movement from combined on-chain and technical indicators — testing whether blockchain-native signals such as investor profitability ratios and liquidity flows add predictive value over price history alone.

Collected, synchronised and cleaned large-scale on-chain datasets from multiple blockchain analytics sources. Engineered and selected features from a literature review plus exploratory analysis, then trained and validated LSTM models against price-only baselines using standard forecasting metrics — the point being the comparison, not the headline forecast.

• LSTM
• Time-series
• Feature engineering
• TensorFlow / PyTorch
• Pandas

Gym Management System

Oracle Apex SQL · database design · agile (scrum / kanban)

A fully functional gym management system built from scratch: use-case diagram, entity relationship model, then normalisation of that ERD to guarantee data integrity and remove redundancy.

99/100 Awarded for the database design and in-class presentation.

• Oracle Apex
• SQL
• ERD & normalisation
• Agile

Inventory Management Application

Java · object-oriented design

A stock management application applying OOP principles for modular, maintainable code: a menu-driven interface with input validation, buy / sell / reprice operations and total inventory valuation.

97/100 Awarded, with credit for exception handling and robustness under bad input.

• Java
• OOP
• Data structures
• Exception handling