Open-Source AI for Africa · Nairobi

AI is already in Africa.
Nobody checked if it works.

Makini AI builds open evaluation tools and language resources so AI works safely and fairly for African communities, in their own languages. Everything we build is open-source and free.

546
Acts of Kenya, open
63,000+
Legal text chunks
1,000+
Iroh users, week one
0.92 F1
Swahili hate-speech
vs GPT-4o at 0.75

Our commitment

Open by default

Every dataset, model, and tool we build is released free and open-source. Our work is a public good: accessible to researchers, developers, governments, and communities across Africa, with no proprietary lock-in.

The stakes are continental. AI could add up to $2.9 trillion to Africa's economy by 2030, transforming agriculture, healthcare, education, financial services, and government. Adoption is already racing ahead of the infrastructure meant to support it. In Kenya, 42% of internet users were using ChatGPT by mid-2025, among the highest rates in Africa, and across the continent AI uptake is surging fastest precisely where people are underserved by Western platforms. But this AI was built, trained, and safety-tested in English and a handful of global languages. The communities adopting it most eagerly are using systems that have never been evaluated in the languages they actually speak. That gap is the difference between AI that empowers and AI that quietly misinforms, on legal rights, on health, on public services. We build the open tools to close it.

Projects

What we're building

Iroh

Live

An AI legal assistant grounded in the complete Laws of Kenya. Kenya's official legal platform is often down, and general AI models hallucinate Kenyan law and procedure, failing the citizens and paralegals least able to catch the errors. Iroh gives accurate, source-cited answers.

1,000+ users in week one
Try on Hugging Face

Kenya Law Corpus

Dataset

An open corpus of the complete Laws of Kenya: 546 Acts and thousands of court judgments, structured into 63,000+ searchable chunks. The substrate for Iroh and for legal-AI research.

View on Hugging Face

MAKINI Bias Detection

Model Dataset

Detecting social and gender bias in Swahili and African French. Built on AfricaBias-SW-FR, the first large-scale African-language bias dataset (35,285 annotated sentences), developed through the AfriLabs AI Accelerator with support from the Gates Foundation.

0.979 F1 macro · +44.7 over nearest baseline · +52.3 over Afro-XLMR

MAKINI Hate Speech

Model

Hate-speech detection for Swahili and French, documenting where models trained on European data fail on African content.

0.92 Swahili F1, outperforms GPT-4o (0.75)
View on Hugging Face

Africa Employment Law

Beta

An AI assistant for employment and labour law across 30 African countries. Workers, HR teams, and legal practitioners get accurate, jurisdiction-aware answers on hiring, contracts, termination, and workplace rights, grounded in each country's labour legislation. Currently in beta.

Try on Hugging Face

Rada

In development

Civic-legal AI for Kenyan languages, helping citizens access governance and legal information in their mother tongue.

Get in touch

Let's work together

We partner with researchers, governments, and organizations building AI that serves African communities.

Nairobi, Kenya