Research
Utafiti Labs — solving complex ML problems in the open.
Utafiti Labs is Kervalt's research arm. We publish, open-source, and peer-review our work on multilingual modeling, retrieval, and agentic reasoning — because sovereign AI should be built on science anyone can inspect.
Programs
Open science, structured for impact
Aya Open
A multilingual open-science model ecosystem: open weights, community-built datasets covering dozens of languages, and transparent governance that lets the world audit and extend our work.
Explore the ecosystemOpen Science Community
A global network of researchers and engineers collaborating on datasets, evaluations, and reproducible methods — with Kervalt providing compute, tooling, and editorial support.
Scholars Program
Residencies and fellowships for researchers tackling hard problems in sovereign AI. Scholars publish openly while working alongside Kervalt's core research teams.
Catalyst Grant Program
Funding and compute grants for independent labs and academic teams whose work advances open, sovereign machine learning — no strings, no lock-in.
Research Areas
Where we push the frontier
Multilingual modeling
Pre-training and alignment techniques that serve dozens of languages equitably — not just the high-resource few that dominate today's benchmarks.
Retrieval architectures
Grounded generation over private corpora: retrieval systems that keep sensitive data inside your perimeter while raising factual accuracy.
Efficient agentic reasoning
Smaller, faster models that plan, call tools, and verify their own work — engineered for deployment in regulated, resource-conscious environments.
Benchmarks
Measured in the open
Every Utafiti Labs release ships with reproducible evaluation suites and public leaderboard standings. We compete on multilingual quality, retrieval precision, and reasoning efficiency — and we publish the harnesses so you can verify the numbers yourself.
40+
Languages covered by open evaluations
100%
Evaluation harnesses published and reproducible
3
Public leaderboards maintained by Utafiti Labs
Ready to deploy sovereign AI?
Run Kervalt models in your own cloud, on your own terms.