Certification
Proof that an AI product truly works in your language.
Evaluation
Planned
Pilots
Process
~/nzinga
$
test sets, then native-speaker review
✓ FLORES+ devtest 1,012 sentences
→ native-speaker review planned
● certificate on a pass
Nzinga · evaluation
languages
Languages measured today
12
test.sets
Open test sets
FLORES+, NTREX-128
reviewers
Native-speaker reviewers per language
Planned
Overview
Nzinga Certified will tell procurement teams whether an AI product works in an African language before they commit. Native-speaker reviewers score real tasks against open test sets, and we publish the results and method.
What it checks
understanding
task
Accuracy on everyday questions and requests
fluency
task
Natural, correct replies in the language
tokens
cost
Token premium against English, measured on FLORES+
safety
review
Harmful or misleading answers tested by native speakers
monitoring
ongoing
Re-tested after every major model update
Process
01
Agree the use case
You tell us the tasks, languages and users the product must serve.
02
Native speakers test
Paid reviewers score real tasks against open test sets.
03
Results are published
Pass, and your product may carry the Nzinga Certified mark.
Ask about a pilot for your product.
Tell us the product, languages and users. We will scope a first evaluation with you.
→ agree tasks and languages
→ native-speaker review
→ publish results and method
✓ open test sets · ✓ paid reviewers · ✓ public method
● planned · pilots open