Portfolio
A collection of hands-on engineering work spanning UAV systems, machine learning, and embedded hardware, built with curiosity, documented with care.
Developed a computer vision dataset for colored vehicle detection and classification using aerial imagery captured with a DJI Mini drone. Gathered over 5,000 images of vehicles from different colors from parking areas within the university campus, producing a well-annotated dataset for object detection and machine learning research.
Built and configured a quadcopter UAV using the F450 frame. Involved soldering, wiring, frame integration, telemetry & GPS setup, calibration on Mission Planner, and iterative flight testing until stable flight was achieved. Concluded with payload integration for agricultural spraying operations.
Assembly and testing of a hexacopter UAV designed for heavier payload operations and improved flight stability. The six-rotor architecture introduced additional complexity in balancing, configuration, and power management. Included payload testing, stability verification, and continual battery monitoring throughout open-field tests.
A real-time electric submersible pump condition monitoring system using ESP32-based current sensing and machine learning to detect faults. Handled the embedded systems build, live data acquisition on the edge, dataset preparation, and contributed to the baseline ML model. Key challenge: reliable sensor data collection at the edge through deep datasheet and communication protocol study.
Developed for the She Code Africa ML/AI Hackathon 2025. Predicted the Hamilton Depression Rating at 6 months using demographic, medical, birth complication, and social support data. Led the ML workflow — building, comparing, and optimizing multiple models with explainability (XAI) built in to keep the user central to every decision.
Carried out as part of the Deep Learning Indaba Community Challenge 2025. Built an AI/ML system to predict the probability of national grid collapse from historical data. Led model development and dataset cleaning, and pitched the idea to the Deep Learning Community Challenge Team — project was acknowledged as innovative.
Deep learning prototype for recognizing handwritten characters using the MNIST dataset, serving as a proof of concept for an OCR system. Explored PyTorch workflows: data loading, image transformations, tensor operations, and neural network design with Flatten, Softmax, and Pooling layers. Careful hyperparameter tuning resolved transformation-related scope issues.
Leveraged AWS AutoGluon to predict bike-sharing demand using Kaggle tabular data. Workflow covered data cleaning, feature engineering, model training, and hyperparameter optimization across multiple iterations — improving prediction accuracy at each pass.
Trained a machine learning classifier to predict heart disease risk from the UCI Heart Disease Dataset (Kaggle). Focused on a genuine need to support the healthcare sector — cleaned the data, explored key features, and built classification models to support early diagnosis workflows.
Built a deep learning model to classify landmarks from images — first with a raw CNN, then with transfer learning using pretrained models. A comparative study in from-scratch training vs. pretrained feature extraction, documenting performance gaps and design decisions at each stage.
Built an image classification model on AWS SageMaker to detect the type of vehicle a delivery driver uses — routing bicyclists to nearby orders and motorcyclists to farther ones. An end-to-end ML workflow from data preparation through deployment for Scones Unlimited's logistics optimization.
Combined embedded systems, mechanical design, and solar power to improve waste disposal. Features automated lid control, level sensing, GSM alerts, and a motor-driven compression system using time-based control. Led a team of 7 from ideation through final presentation, ensuring every component of the project pipeline came to life.
Built a system to detect power availability and WiFi signal strength at multiple campus locations in a single device — giving students real-time access to building power status via an app. Led the team from ideation to final presentation. Outcome: learned dashboard design and user-first thinking in engineering.
A collection of Arduino simulation projects on Wokwi to test logic systems, embedded workflows, and hardware behavior virtually before physical implementation. Each simulation required critical thinking about circuit design, component interaction, and control logic.
A conceptual embedded systems project tackling parking lot management through automation and sensing. Implemented through simulation with all real-world components represented — demonstrating how automated detection and occupancy logic could remove stress from the parking experience.