My work includes AkiliCode, a code-focused model initiative, and Sauti, a family of Swahili voice models comprising Sauti ASR and Sauti TTS.
I'm currently pursuing an MSc in Financial Engineering at WorldQuant University, with research interests in reinforcement learning for sequential decision-making problems in finance.
I also write about multilingual language models, African NLP, and applied machine learning.
Lessons from bringing Swahili voice AI to farmers in Busia and Ahero through ordinary phone calls using only airtime, and what their response taught us about trust, reliability, and access.
Sauti-Dia-SW is a full fine-tune of Dia-1.6B trained on 126 hours of Swahili speech from a 500-hour corpus we assembled from 15 openly licensed public datasets. Released under CC-BY-4.0, with model weights on Hugging Face
We partnered with Africa Compute Fund to explore collaboration on AI infrastructure, deployment, distribution, and commercialization, helping accelerate the adoption of African-built AI.
As Africa advances in building AI systems that support its many languages, a deeper challenge emerges: understanding meaning beyond words. True AI inclusion requires more than linguistic capability. It demands cultural awareness, cognitive flexibility, and alignment with African values and ways of reasoning. This article argues that language is only the entry point. To build AI that genuinely serves African communities, we must consider culture as context, cognition as structure, and Ubuntu as a guiding design principle. Ultimately, Africa's most significant contribution to the AI era may not be new models or datasets, but new perspectives on what intelligence should understand, value, and serve.
MsingiAI Joins NVIDIA Inception Program
Get an email whenever I publish a new article.
I'm an AI researcher and engineer based in Nairobi, Kenya, and a co-founder of MsingiAI, where we build foundational AI models for African languages. My work focuses on African-language NLP, speech technologies, and agentic AI systems.
At MsingiAI, I work on AkiliCode, our code-focused model initiative, and Sauti, our family of Swahili voice models. Sauti includes both Sauti ASR (automatic speech recognition) and Sauti TTS (text-to-speech), designed to make high-quality Swahili speech technology openly accessible.
I'm also an active contributor to the African AI research community. I write about multilingual LLM safety, Swahili benchmarks, and the current state of African NLP across platforms like LinkedIn and the MsingiAI research blog. I've also spoken at events such as a panel hosted by Microsoft Research Africa, discussing practical challenges and opportunities in deploying African-language voice AI.
I'm currently pursuing an MSc in Financial Engineering at WorldQuant University, where my research interests include reinforcement learning applied to market-making, optimal execution, and multi-asset portfolio control.
I strongly advocate for open, community-driven AI that serves the 1.4+ billion Africans whose languages remain underrepresented in today's models.
A lightweight library for detecting data drift in machine learning models, ensuring production model reliability and performance monitoring.
A Kenyan Swahili Voice Model designed to democratize speech technology for local languages.
A Python library that wraps around any prediction function or ML model to automatically attach uncertainty quantification methods.