Language data
Dataset provenance, quality review, representation, and licensing.
Researching multilingual language intelligence, model adaptation, evaluation, and responsible AI deployment for African contexts.
Our proposed first research milestone is to compare open-weight multilingual models, curate legally usable Somali-English data, run controlled adaptation experiments, and publish measured findings.
Success will be judged against held-out evaluation tasks covering fluency, instruction following, factual reliability, and safety—not by training completion alone.
No released Lubito-1 model or independently verified performance claims at this stage.
Dataset provenance, quality review, representation, and licensing.
Human evaluation, multilingual benchmarks, regression and safety testing.
Efficient inference, private hosting options, and operational controls.