Soon
AfroEval
Test your agent in Pidgin, Yoruba, Hausa, Igbo, Swahili and Twi before your customers do.
The split
- LLM writes
- Proposes candidate tasks for native speakers to verify
- System One decides
- Grades answers against rubrics
- Code executes
- Runs suites, repeats, computes cost and accuracy
Engine
Task suites written with native speakers; repeated runs for consistency; cost tracked per task, so results read as accuracy per dollar.
Rulebook
What correct means here, and the mistakes we've already caught. It grows one job at a time.
- R01Native speakers write and verify every task.
- R02Report accuracy per dollar, not just accuracy.
- R03Publish the method; keep test sets private so nobody trains on them.
Intake
- Agent endpoint
- Target languages
- Task types
- UNIT
- One agent evaluated across a language suite
- REVIEW LAYER
- Native-speaker reviewers score a sample of every run.
- DELIVERY
- A scorecard per language and task, with failing examples.
- TURNAROUND
- 1–2 weeks
How we find buyers
- Model labs and startups expanding into Africa
- Language coalition partners
- Published results via Griot
More like this
LiveResearch
Griot
Field notes on AI in Africa: research, teardowns and builder guides, published weekly.
- UNIT
- One issue per week; sponsor slots
- PRICE
- Free · sponsor slots from $300
WeeklyOpen
BetaResearch
Ecosystem Intelligence Reports
Sourced reports and datasets on who is building what with AI in Africa, for ecosystems, funds and foundations.
- UNIT
- One report with its dataset
- PRICE
- From $2,500 per report
2–3 weeksOpen