In recent years, Artificial Intelligence has aided in the development of many solutions around the world in numerous fields and it has become imperative to ask how advanced AI methods like Machine Learning and Natural Language Processing can be reconstructed so as to make the Internet more inclusive practically, for communities in low-resource environments in the Global South. This paper outlines the project in Ghana, which utilizes a Participatory Action Research approach to build a local voice-based Automatic Speech Recognition system to provide domain-focused web-based information to local communities in Dagbani (a local Ghanaian language). We look at the methodology, and how it utilized insights from community engagement to build an inclusive system. We also look at what broader implications of this design process for the Web and AI in the context of decolonization of the Internet.