Keza AI
Preparing your celebration…
Select Language
Celebrate Every Moment with Keza AI
Keza AI
Preparing your celebration…
An AI stating "this is how Rwandans do X" is a factual claim, not flavour text. So instead of trusting model output alone, Keza AI runs a deliberate governance pipeline.
01
A cultural claim enters as an unverified draft, never shown to users in that state.
02
Claims are checked against independent sources before anyone signs off on them.
03
A named reviewer with cultural context approves publication; no auto-publish path exists.
04
Live in Keza & Ndolo; a report button lets any user flag an error into a moderation queue.
The pipeline has already caught a real error: a prior release had mislabelled a Kikuyu (Kenyan) term as Rwandan practice, exactly the kind of mistake this process exists to catch. It's now fixed, sourced, and the same governance layer covers all 22 countries in the core assistant and 21 in Ndolo, each entry carrying its source, reviewer, and a confidence rating.
A cultural practice is almost never true 100% of the time for everyone in a country. Every claim carries an explicit confidence level instead of manufactured certainty.
Established
Widely confirmed across multiple independent sources and reviewers.
Common practice
Broadly true, with acknowledged regional or generational variation.
Contested
Sources disagree. Shown as a disagreement, not resolved by guessing.
Unverified
Never shown to users: the state every claim starts in.
Every Keza and Ndolo reply can be flagged directly in the conversation. Have cultural expertise to bring beyond a one-off correction? Get in touch.
Contribute cultural knowledge