Engineering an Afrocentric AI Platform: A Foundational Architecture for Training and Renewal

Abstract The paper explores the need to design a systematically engineered technical architecture for developing and incorporating Artificial Intelligence (AI) models, especially Large Language Models (LLMs), in the socio-technical environment specific to Africa. Although AI's potential to transform the continent is boundless, serious gaps remain in systematically engineering datasets from varied, frequently informal, and fractured African sources for successful AI model ingestion. Traditional AI approaches, especially in data engineering and model training, tend to be inadequate or biased when adopted without profound adaptation to the African environment. Therefore, a design philosophy prioritising local relevance, cultural specifics, and direct engagement with regional issues is urgent.It focuses on the engineering perspective needed to provide the basis and approach for designing and implementing foundational data architecture to fill existing gaps. Similarly, the proposed Afrocentric AI Training and Renewal Architecture introduces a unified framework for the smooth incorporation of data from academic libraries, transaction management systems, and activities in informal processes into an autonomous, self-renewing AI system. This architecture, spanning from data sourcing and engineering to training and deployment with a closed feedback loop, aims to produce culturally relevant and efficient AI solutions for the continent. The success of the resulting engineering solution rests on interdisciplinary potentials, which focus on social sciences, economics, and cultural studies to make the resulting datasets highly aligned with the socio-economic realities across the continent and genuinely fit-for-purpose.