Luke Ross

Endangered Species EDA

• Utilised SQL to extract data from 10 different related tables from CITES Trade Database using JOIN and VIEW
• Transformed and filtered data in SQL by using aggregating & filtering function to improve reporting process
• Loaded and visualised data with Python identifying key trends & objectives for the conservation of endangered animals
• Developed interactive dashboards in Power BI that provided insights to support decision-making in conservation planning

Pharmaceutical Price Analysis Around the World

• Extracted pharmaceutical pricing data across multiple global datasets using SQL (JOIN, FILTER, AGGREGATE)
• Cleaned and standardised inconsistent pricing records to improve comparability
• Built interactive dashboards in Power BI to visualise cost disparities across countries and support healthcare policy discussions

Amazon Product Web Scraper & EDA

• Scraped & ingested product data using Python (BeautifulSoup) & cleaned data was centralised in Azure Synapse Analytics
• Stored & managed the large datasets in Azure Synapse Analytics building ETL pipelines to clean, aggregate & standardise data
• Enabled SQL queries through Microsoft Azure & connected Azure Synapse Analytics to Power BI for creating dashboards
• Interactive dashboards were created in Power BI as well as further exploratory data analysis in Python & RStudio

Data Professional Survey Dashboard

• Retrieved raw survey data & imported into Power BI for ETL
• Cleaned the data involved standardisation of job title names, handling missing values, outlier & filtering inconsistent entries
• Data was restructured via Power Query where columns were split, merging & categorical variables were reformatted
• Meaningful DAX measures were built and a dynamic dashboard was developed where stakeholders could explore by role, salary band & geography

Lego Models EDA

• Collected and processed LEGO datasets using SQL queries and Python scripts
• Analysed historical release trends, popular themes, and part diversity to uncover consumer and design patterns.
• Applied DAX measures and data modelling to track correlations between number of pieces, retail pricing, and release year.
• Developed interactive dashboards in Power BI to identify KPIs and provide data-driven insight

SpaceX Flight Landing Prediction

• Collected launch data by scraping SpaceX launch records from Wikipedia with BeautifulSoup and integrating the SpaceX REST API
• Performed ETL, cleaned the data & developed an interactive geospatial maps of launch sites with Folium
• Various models trained such as logistic regression, decision trees, SVM & KNN
• Developed an interactive dashboard in Plotly Dash that displays launch‑success statistics, payload/booster correlations and allows filtering by launch site
• Leveraged Python/Jupyter along with Pandas, NumPy, BeautifulSoup, Requests, Scikit‑learn, Plotly, Folium and Dash across the project

Mathematics for Machine Learning Project

• Examined vector spaces, orthogonality, projections and inner products & angles
• Developed notebooks which utilised inear algebra, machine learning algorithms and neural network theory
• Implemented algorithms such as K‑Nearest Neighbours (KNN) and PageRank to translate theoretical insights into ML code
• Practised gradient-based optimisation and backpropagation on problems involving fitting probability distributions and modelling helices with neural networks.

Deep Learning Project

• Explored a range of deep‑learning architectures through experiments with Convolutional Neural Networks (CNNs), Vision Transformers (ViTs) for a wast classification problem
• Built and compared CNN classifiers in Keras and PyTorch, including a custom PyTorch model trained from scratch with comprehensive training/validation monitoring
• Implemented Vision Transformers in both frameworks, experimented with hybrid CNN‑ViT architectures and analysed their training efficiency relative to CNNs
• Developed predictors for breast‑cancer classification, investigating memory‑based versus generator‑based data pipelines and using cross‑validation for evaluation
• Applied transfer learning and fine‑tuning, with model selection and hyper‑parameter tuning for waste image classification

Financial Fraud Detector

Generative Pre-Trained Transformer (GPT)

• Developed a GPT utilising self-attention, layer norms, embeddings & residuals using PyTorch in AWS SageMaker & Docker replicating the 124-million-parameter GPT-2 model
• Data pipelines produced with tiktoken; used multiprocessing developed for tokenisation, sharding & ingesting datasets
• Tuned models using cross-entropy loss & optimised with cosine learning-rate scheduling, dropout & mixed precision acceleration

Real-Time Facial Muscle Activation Detector for Locked-In Syndrome

• Signal processing techniques such as Chebyshev, Butterworth filters & wavelet transforms performed in MATLAB & Python
• Trained a binary classifier in AWS SageMaker using scikit-learn & XGBoost, achieving 92.3% accuracy, 100% precision & 85.71% recall on unseen data
• Designed end-to-end ML pipeline: data ingestion, preprocessing, model training, real-time prediction & reproducible via Docker

Churn Model

Computer Vision

Churn Model

PK/PD Modelling to Assess The Effects of Anti-Cancer Agents on Tumour Volume

• In collaboration with GlaxoSmithKline, MATLAB, Python & Excel was utilised to forecast an optimal dosing strategy
• Dataset was transformed, cleaned & standardised performing missing value imputation, outlier detection & normalisation
• Model parameters were validated, fitted, visualised & then results were reported to stakeholders
• Followed compliance and carried out risk assessments for working with sensitive data
• A novel strategy was produced for combined therapies, and the model was stress tested

Simulation-Based Mathematical Models & Parameter Estimation to Interpret Tissue Growth Experiments

• Studied cell migration in tissues using stochastic lattice‑based simulations, PDE continuum approximations and discrete agent‑based random walks
• Compared measurement‑error models by fitting simulation outputs to experimental‑style data, evaluating additive Gaussian versus multinomial error formulations for single & multiple subpopulation scenarios
• Provided synthetic data and scripts for parameter estimation, identifiability studies and prediction interval calculations
• Heavy reliance on numerical methods, statistical modelling PDE solving & inference

StochLab: Interactive Stochastic Modelling Dashboard

• Developed a full-stack interactive dashboard utilising Python & REST API to simulate stochastic processes & financial models
• Developed numerical methods in Python for on-demand parameter exploration & statistical modelling within the dashboard
• Built Monte Carlo, Gillespie, Brownian models in Python, deployed on GitHub & used Docker containers for reproducibility