What actually moves the pipeline forward.
AfriFoundry works with universities, NGOs, language holders, and open-data initiatives — real sources only, never scraped and fabricated to look complete.
Language corpora — the biggest gap today
Text and audio in Swahili, Kikuyu, Dholuo, Kalenjin, Maasai, Somali, Suba, Kambe, Luhya, and more. Parallel text (the same content in two languages) and paired audio-with-transcript are the most valuable — they're what the dictionary researcher and voice model actually need.
Real commerce and market data
Pricing, listings, and vendor patterns from informal and secondhand markets specifically — the segment formal retail data doesn't cover.
Sector-specific data
Real feasibility studies, market reports, and outcomes per sector — what lets Afri3B push back with something real instead of a generic guess.
Cultural context, as it's actually used
Real writing that contains cultural terms in real use — never generated or inferred from general knowledge.
What we won't take: anything without clear provenance, or data requiring consent we can't verify. The same non-fabrication discipline that governs every response governs what goes into training it.
Know a language Afri3B is still learning? Teach it, live.
A mini version of Afri3B lives right here, connected to the training pipeline. Teach it a word, correct something it got wrong, or just talk to it — every message is reviewed by hand before it enters the self-building dictionary, the same honesty discipline as everything else here.
Before you start
Anything sent here may be reviewed and used to train Afri3B. Don't share personal or private information about yourself or anyone else, or text you don't have the right to share. By continuing, you're confirming this is your own knowledge and you're comfortable with AfriFoundry using it to train Afri3B, per our Privacy Policy and Terms.