The lack of publicly available autonomous vehicle datasets from developing regions—particularly across diverse African road environments including urban, rural, and unpaved terrain—hampers progress in robust perception for low-resource settings. We introduce a procedural augmentation pipeline that enriches low-cost monocular dashcam footage with realistic refractive distortions and weather-induced artifacts tailored to these challenging African driving scenarios. Our refractive module simulates distortions from low-quality lenses and air turbulence, including lens distortion, Perlin noise, Thin-Plate Spline (TPS), and divergence-free (incompressible) warps. The weather module adds homogeneous fog, heterogeneous fog, and lens flare. To support autonomous perception research in underrepresented African contexts—without the need for costly data collection, labeling, or simulation—we release our distortion toolkit and augmented dataset splits along with the baseline restoration model.

An investigation into testing practices and methodologies for Jupyter Notebooks, exploring ways to improve code quality and reliability in data science workflows.

A comprehensive deep learning pipeline for object classification and camera alignment on Mars rover imagery. The project implements advanced semantic segmentation models including DeepLabv3, LRASPP, and SAM for cross-domain object detection. Key contributions include the development of custom data loaders, unified label systems, and extensive data cleaning and augmentation pipelines. The work also incorporates SLAM techniques for 3D reconstruction and utilizes Infinigen for synthetic data generation to enhance model training.

This project explores document retrieval techniques by implementing and comparing sparse, dense, and hybrid methods. Using BM25, Sentence-BERT, and FAISS for retrieval, we evaluate their effectiveness on the MS MARCO dataset. Performance is measured using Mean Reciprocal Rank (MRR) and retrieval time. The project highlights the strengths and limitations of each approach and examines how combining sparse and dense methods can improve retrieval quality.