Selected Work — CS / DS / ML / AI
A multi-layer perceptron built from scratch in PyTorch, trained on tabular datasets with custom regularisation techniques including dropout and L2 penalty. Achieved 91% accuracy on held-out test sets across multiple benchmark datasets.
Fine-tuned a BERT-based transformer model for financial news sentiment analysis. Built a full preprocessing and inference pipeline with HuggingFace Transformers, achieving 94% accuracy on a labelled corpus of 50,000 articles.
An interactive web dashboard exploring UK economic indicators across five decades of data. Built with React and D3.js, featuring animated transitions, responsive charts, and a custom colour-coded timeline interface.
A Python framework for backtesting momentum-based trading strategies on historical equity data. Includes Sharpe ratio analysis, drawdown metrics, and a visual performance report generated automatically on each run.
Ensemble regression model predicting London housing prices using public datasets from the ONS and Land Registry. Combined gradient boosting with feature engineering across postcode-level socioeconomic indicators.
A constraint satisfaction solver that generates optimal weekly timetables for LSE students, minimising clashes and travel time between buildings. Built using OR-Tools with a simple web interface for input and output.