Projects & Portfolio

Deployed Streamlit app from my MSc thesis, serving three trained recommender models live: matrix factorisation, CatBoost, and feature-augmented LightGCN. Offers top-K and cold-start recommendations, model comparison, and per-recommendation explanations.

Empirical comparison of seven recommender variants across three model families on MovieLens-1M (1M ratings) under a single leakage-free protocol. Separated loss-function effects from architecture and validated every headline result with paired-bootstrap significance testing.

Dynamic BI dashboard analysing £2.30M in sales and £286K in profits. Implemented drill-down functionality revealing regional trends, optimal discount strategies, and seasonal peaks for marketing decisions.

Comprehensive Power BI dashboard analysing global electronics sales. Identified top-performing categories, seasonal spikes, and online vs in-store trends to inform strategic growth opportunities.

Robust book recommendation system using collaborative filtering and transformer architectures. Performed thorough quality checks on Bookcrossing dataset, refined through model experimentation.

Predictive model assessing term deposit likelihood for bank clients. Compared Logistic Regression, XGBoost, MLP, SVM, and Random Forest using stratified K-Fold validation and dimensionality reduction.

Six-week sales forecast for 1,100+ Rossmann stores across Germany. Engineered temporal features for promotional periods and seasonal trends, enhancing inventory and staffing decisions.

Machine learning recommendation system for Zimnat analysing 40,000+ customers. Devised reliable cross-selling predictions based on demographic factors and purchasing behaviour patterns.

Trend analysis of WeRateDog Twitter dataset. Corrected data quality issues, executed detailed transformations and visualisations to uncover meaningful social media engagement patterns.

Investigation of student performance factors. Extracted 29 key indicators from 600+ features, employed multivariate visualisation to examine parental engagement effects on outcomes.

Exploration of comprehensive movies database trends. Drafted eight research questions, performed extensive cleaning, and presented results through diverse visualisations yielding in-depth insights.

Scrapy-based crawler for streamlined Punch news collection. Captures titles, content, and links with Scrapyhub deployment capability, storing data in multiple formats for analysis.

Scrapy framework gathering user-submitted names from behindthename.com worldwide. Expanded from Eastern Africa to all countries, compiling structured dataset of cultural naming trends.

Examination of water usage and tap flow rates across locations. Delivered flexible module-based framework supporting evidence-based decision-making in resource management and conservation.