Projects
A selection of data science, ML, and AI projects — from regulatory financial analysis and drug discovery to personal experiments with data. Code available on request.
Analysing Quarterly Results of Globally Systemically Important Banks with Advanced Language Models
Led a cross-functional team of 5 learners on a 6-week project in partnership with the Bank of England (Prudential Regulation Authority) and the University of Cambridge. Applied advanced language models to quarterly earnings call transcripts of financial institutions to proactively assess firms' Financial Stability and Risk, proposing a novel Score (FSRS) — translating qualitative supervisory signals into a single quantitative firm-level indicator.
Built the underlying NLP pipeline (sentiment, topic modelling, linguistic framework) and designed a RAG-based query system over 5 years of transcripts using LangChain, ChromaDB, Phi-3, and Gemini Flash. Designed an LLM selection framework and a bias-mitigating data sampling strategy. Presented to senior data scientists at the Bank, delivering the codebase and final report.
Book Sales Forecasting with Time Series and ML
Data-driven inventory and marketing decisions
Forecasts book sales using time series data and advanced machine learning models — including ARIMA, XGBoost, LSTM, and CNN — with a full end-to-end MLOps pipeline covering production deployment.
AI Driven Nanobody Discovery for SARS-CoV-2
Accelerating drug development through intelligent protein analysis
Leverages advanced AI techniques to accelerate nanobody development for SARS-CoV-2, focusing on fine-tuning protein language models, predicting nanobody-antigen binding affinity, and benchmarking performance metrics across generative model architectures.
Ground Truths: What My Coffee Data Revealed
This project applies data science principles to a hands-on tasting experiment aimed at challenging a personal assumption: that I dislike light roast coffee.
I investigate roast preferences, brewing methods, and variability in how I rated each cup, turning daily coffee into a data-driven discovery.
Customer Segmentation with Clustering
Data-driven insights for retail success
Applies machine learning to design clustering models for customer segmentation, enabling smarter marketing strategies and improving customer experiences through data-driven persona development.
Detecting Anomalous Ship Engine Activity
Ensuring optimal maritime performance through intelligent monitoring
Tackles anomaly detection in a no-ground-truth setting to ensure optimal ship engine performance. Applied statistical and ML-based approaches including one-class SVM and Isolation Forest to flag anomalous engine behaviour.
UCL Cancer Hackathon
Awarded joint first place at the UCL Cancer Hackathon. Developed a specialised data pipeline for personalised cancer vaccine design, targeting high-efficacy neoantigens. The solution uses genomic data and T-cell receptor (TCR) structure to filter out ineffective targets, aiming to improve patient response rates.
Note: projects on student dropout prediction and a popular gym chain are omitted due to a confidentiality clause preventing discussion of results or sharing of data. Cleaned repo code is available on request.