Congratulations on your win! Could you give an overview of your winning innovation?
MansaLLM is a language model engineered specifically to understand, translate, and reason across African languages. Unlike many global LLMs that struggle with low-resource languages or treat them as afterthoughts, MansaLLM is fundamentally designed to operate in African linguistic contexts with appropriately aligned cultural and contextual depth.
MansaLLM’s architecture combines several innovations. First, the model is trained on high-quality, culturally grounded, ethically sourced multimodal datasets spanning speech and text, across monolingual and multilingual settings, tailored to African communicative norms.
Second, to ensure accuracy and cultural fidelity, MansaLLM undergoes a structured pipeline that includes human-in-the-loop evaluation with native speakers and continuous learning loops that refine performance over time. Machines draft. Humans decide. That’s our fidelity loop.
Finally, MansaLLM supports contextual translation, transcription, localized content generation, and interactive communication tools for real-world applications from enterprise workflows to consumer tools.
All three innovations are key. Unlike generic LLMs that predominantly focus on European and Asian languages, MansaLLM is tuned for African languages and contexts; emphasizes cultural and contextual intelligence, capturing idioms, tonal nuances, and complex morphology unique to African languages; and is part of a multimodal ecosystem that goes beyond text to include spoken and visual understanding in future versions.
MansaLLM currently supports more than 20 African languages for translation and localization tasks in production systems, with broader support for 10 additional languages in beta. The 20 languages are
Afrikaans, Amharic,
Arabic, Chichewa, Hausa, Igbo,
Kinyarwanda, Luganda, Malagasy,
Oromo, Sepedi,
Sesotho, Shona,
Somali, Swahili, Tswana, Tsonga, Xhosa, Yoruba, and
Zulu. The 10 in beta mode are Bambara, Bemba, Ewe, Kikongo, Kirundi, Lingala, Ndebele (South), Swati,
Tigrinya, and Twi.
Could you give us an overview of the African Languages Lab and the work you do through it?
Broadly speaking, the African Languages Lab, as a social enterprise, is both a research initiative and an operational engine for advancing
natural language processing (NLP) and AI adoption across African languages. It was established to systematically address the digital exclusion of Africa’s linguistic heritage, leveraging research, engineering, and community engagement to build AI that scales, for Africans by Africans. We conduct research that builds tools, and tools that build communities.
To get a bit more detailed, we create, assemble, and clean linguistic data from diverse sources, including spoken corpora, text archives, and community contributions, creating some of the largest multimodal language resources for Africa.
We also mentor researchers and foster collaboration among universities, developers, and language communities to sustain local expertise. Those relationships help in publishing research and integrating with global platforms, allowing us to assist organizations in incorporating African language AI into global workflows.
Did you run into any development issues when creating MansaLLM?
Developing MansaLLM was not without setbacks. The principal challenge is the lack of robust, consistent datasets, especially for languages with primarily oral traditions or scant written corpora.
In addition to creating large amounts of our data in-house, using our custom tools such as All Voices, we also engaged native speakers, linguists, and institutions to source, clean, and validate data. This yielded expansive multimodal datasets covering thousands of hours of speech and billions of tokens of text.
Instead of relying solely on automated benchmarks, we integrated human-in-the-loop evaluations to ensure linguistic nuance and cultural relevance. MansaLLM also uses pipelines that allow ongoing improvement as more data becomes available, mitigating initial sparsity issues.
These efforts helped close performance gaps with high-resource languages and even achieve competitive or superior accuracy on several African language benchmarks compared with generic translation tools.
Can you detail how MansaLLM is designed to enhance linguistic efficiency rather than replacing the linguist?
Our philosophy is rooted in an Ubuntu approach to human-centric AI: equipping linguists and creators with tools that automate repetitive, first-pass, and time-consuming tasks, like draft translation, alignment, and transcription, while allowing humans to retain control over nuance, tone, and cultural authenticity.
In practice, this means: draft translations and first-pass generations that experts can review and refine; leveraging of glossaries, terminology management, and contextualized style supports that help teams maintain uniform messaging without starting from scratch; and interfaces in which human translators and AI can iterate together.
The result is increased productivity and accessibility for linguists serving markets that previously lacked any AI support. Thus, creators are empowered to generate and innovate much faster without sacrificing quality.
Do you see MansaLLM as the start of a larger project that will eventually include more African languages?
Roughly fewer than 3% of the more than 2,000 African languages on the continent have representation in AI systems today — a stark gap that MansaLLM begins to address.
A larger ecosystem has been created not only by the African Languages Lab, but by other commendable initiatives across the continent, largely driven by young brilliant minds with an electric sense of purpose and goals of making big impacts.
For the African Languages Lab, future plans include building solutions to Africa’s most pressing, real-world challenges, ensuring this technology delivers not just technical novelty, but tangible value where it matters most.
Another future priority is enabling community APIs and developer platforms so local technologists can build their own language services. We also want to dedicate tools for oral languages and script-diverse systems that reflect Africa’s linguistic complexity and create a full conversational multimodal language model that spans text, speech, and images.
The ultimate vision is an AI landscape in which every African language, whether major or marginalized, has a seat at the digital table.
What are your observations of the continental market, what does the future look like, and what challenges need to be overcome in regards to African language work?
African language technology is among the most underserved areas in global AI. Most mainstream AI and NLP progress has focused on high-resource languages, leaving vast speaker communities without meaningful tools. This creates real economic and social exclusion.
Investing in African language technologies unlocks access to large, underserved user bases, enabling localized digital services across sectors such as finance, healthcare, and commerce, driving measurable returns and long-term market growth. Multilingual and mother-tongue learning platforms also improve access, comprehension, and outcomes, particularly in early education and workforce training, helping reduce systemic inequities in knowledge access.
Language technologies offer a critical opportunity to preserve culture and linguistic continuity by documenting, preserving, and revitalizing African languages through digital archives, speech technologies, and AI-assisted tools. These efforts promote inclusive technological participation by enabling communities to not only consume technology but also actively shape, adapt, and govern it in ways that reflect local values and realities.
But there are still outstanding challenges, including data scarcity. For many languages, spoken data and orthographies need formalization. Funding and infrastructure are also issues; sustained investment is required to build local expertise and computing resources. And there’s still work to be done in standardization across dialects and scripts.
Overcoming these barriers will require collaboration among governments, universities, communities, and the AI industry. But the momentum, whether through research hubs, startups, or open platforms, is unmistakable.
