SwahiliAI: Building an Open-Source Artificial Intelligence Ecosystem for East African Languages

SwahiliAI is a 24-month research proposal to build an open-source AI ecosystem for Swahili and East African languages. Building on existing African NLP work (Masakhane, IrokoBench, UlizaLlama), the project will (i) create openly licensed, domain-specific Swahili instruction data; (ii) fine-tune open-weight large language models with LoRA/QLoRA; (iii) release SwahiliBench-D, an evaluation suite measuring both accuracy and safety; and (iv) pilot applications on Web, SMS and USSD. The first phase focuses on healthcare, agriculture and education, with extensions to financial inclusion, legal aid and employment. Principal researcher: Rodgers Ochieng Oduor (SPRINTER, Nairobi), in collaboration with Dr. Ibrahim Tirimba Ondabu, KCA University.