Open to new opportunities
Dylan Kennedy
Results-driven Data Analyst at Cartrack, turning data into decisions through Python automation, reporting pipelines, and applied machine learning.
Open to new opportunities
Results-driven Data Analyst at Cartrack, turning data into decisions through Python automation, reporting pipelines, and applied machine learning.
About me
I'm Dylan Kennedy, a results-driven Data Analyst currently at Cartrack, where I design automated Python reporting pipelines, write complex PostgreSQL queries across multiple schemas, and build analytics that drive real business decisions for the customer care team.
I hold both a Bachelor's and an Honours degree in Information Technology from Eduvos, with specialisations spanning Data Science and Robotics. My Honours research produced a legal case outcome prediction system combining LegalBERT with explainable AI — tailored for the South African mixed legal system.
I have a proven track record of building end-to-end ML solutions: from autonomous surveillance robots and medical imaging CNNs to legal AI and classical ML pipelines. Proficient in Python, SQL, and cloud technologies, I'm passionate about leveraging data to solve real-world problems.
Work I'm proud of
Advanced ML model predicting South African legal case outcomes with integrated explainability. Features an outcome classifier, a case similarity engine that surfaces relevant precedents to cut research time, and Explainable AI techniques — bridging complex models with practical legal decision-making.
Raspberry Pi robot that autonomously detects and follows a person using TensorFlow Lite (EfficientDet Lite) via live camera feed. Supports manual override through an Xbox controller. Features PWM motor control for smooth movement and seamless switching between autonomous tracking and manual modes.
Fine-tuned ResNet-50 CNN in PyTorch that classifies brain MRI images into four categories: Glioma, Meningioma, No Tumor, and Pituitary. Evaluated via confusion matrix to identify challenging tumor types and assist earlier, more accurate diagnoses.
CNN built with TensorFlow & Keras trained on 27,000+ cell images to classify parasitized vs. uninfected cells. Used early stopping over 50 epochs, achieving 90%+ accuracy by epoch 25. Binary cross-entropy loss with Adam optimiser.
KNN classifier (tuned via GridSearchCV) trained on 768 patient records to predict diabetes. Achieved 69.70% accuracy and 53.33% F1 score. Includes StandardScaler preprocessing, confusion matrix visualisation, and a live user-input prediction interface.
Solves a Travelling Salesman Problem variant using selection, crossover, and mutation across generations. Evolves optimal station routes from a CSV-loaded dataset, with convergence visualisation and best-path reporting.
K-Means clustering on a 2D dataset with StandardScaler preprocessing. Optimal cluster count determined using both silhouette scores (k 2–15) and the elbow method — converging on 3 clusters, visualised with boundary plots.
Linear regression model (p(g) = 0.11g + 21.71) predicting milk can prices from weight. Trained on 100 entries, achieving R² of 0.91 (train) and 0.92 (test). Predicted a 1.5 kg can at ~R187.70, validating the model's real-world applicability.
Six-servo robotic arm (waist, shoulder, elbow, wrist pitch, wrist rotation, gripper) controlled in real time via keyboard. Built with Python, pyFirmata, and the keyboard library — demonstrates foundational robotics and servo control.
Web development
Responsive marketing website for a Faerie Glen café serving South African and American fusion comfort food. Multi-page build with a downloadable menu and weekly specials, an events and tastings enquiry flow, image gallery, Google Maps location, and social integration.
Business-building agency website spanning home, about, services, and contact pages, with a structured services breakdown and clear strategy-session calls to action. Pairs a static marketing front-end with a separate client login portal.
Brand and marketing site for my own software, data, and machine learning studio. Multi-page editorial layout presenting the three core service lines, with a clean aesthetic and a companion login portal for clients.
Career
Toolbox
Contact
I'm open to full-time roles, contract work, and interesting collaborations in machine learning and data science. Drop me a message and I'll get back to you promptly.