Research Publications
Explore our contributions to the global AI research community, focusing on inclusive technology and African language processing.
Browse by Research Area
Published Research
A Principle of Least Action Approach to Multi-Factor Route Optimization
Traditional GPS navigation systems optimize routes primarily for shortest travel time or shortest distance. While effective in well-mapped urban settings, these simple criteria can lead to suboptimal or even unsafe routes in challenging environments, especially in rural areas with incomplete infrastructure and variable road conditions. We formalize this approach through Thabhelo's Action Functional and establish Thabhelo's Action Routing Principle (TARP) as the theoretical foundation. We propose a novel route optimization framework inspired by the principle of least action from physics. By formulating route planning as the search for a path that minimizes a holistic "action" functional; one that combines multiple cost terms reflecting time, distance, energy, and various hazards; we create a unified, extensible optimization framework. This Lagrangian approach allows custom penalty terms for user-defined concerns, yielding highly flexible navigation suitable for autonomous vehicles in diverse and challenging environments.
Offline-First AI Infrastructure for Emerging Markets
A comprehensive framework for deploying AI systems in low-connectivity environments, enabling reliable AI applications across Africa's diverse infrastructure landscape.
Cultural Bias in Large Language Models: An African Perspective
An empirical study of cultural bias in existing large language models and proposed methods for creating more culturally inclusive AI systems.
Recent Preprints
Scaling Foundation Models for African Languages: A Multi-Domain Approach
We present advances in scaling foundation models across multiple African languages and domains, demonstrating improved performance on downstream tasks.
Federated Learning in Resource-Constrained Environments
A novel federated learning framework designed for environments with limited computational resources and intermittent connectivity.