Sabi
A tutor that reaches the phone a child already has.
Sabi teaches literacy and numeracy over a phone call. A child does not need internet, a smartphone, an app store, a data plan, or a parent who can navigate a web platform.
internet required by the learner
minutes per voice lesson
target annual cost per child at scale
The wedge
The bottleneck is reach.
Nigeria has the highest number of out-of-school children in the world, and millions more are enrolled but unable to read a simple sentence. The usual edtech answer assumes a smartphone, reliable data, and a child with enough quiet time to sit in front of a screen.
Sabi starts from the opposite assumption. Most families can access a basic phone. So the product is a call: short, adaptive, low-bandwidth, and designed around Nigerian speech, market math, and the reality that the learner may share a device with the whole household.
How it works
A phone-call stack, built for accent and cost.
The hard part is not calling an LLM. The hard part is making the whole loop feel patient, audible, and cheap enough for the child who was never in the market for AI tutoring.
Input
Africa's Talking and Twilio demos route the call into a voice loop.
Speech
Whisper listens with Nigerian-English prompt biasing.
Tutor
Claude Haiku runs the adaptive literacy and numeracy lesson.
Memory
Supabase and pgvector keep cross-call learner context.
Voice
Chatterbox TTS gives Sabi a warmer, less robotic voice.
Cost
Self-hosted GPU inference keeps the intervention designed for scale.
Outside diligence
People have already chosen to back it.
- MIT Sloan Africa Business Challenge — 2nd place
- Wesleyan New Venture Awards — Grand Prize
- UNICEF AI Ventures Accelerator — inaugural cohort
Next
Sabi is looking for partners who care about reach, not demos.
If you work on African education, telecommunications, grantmaking, or voice AI infrastructure, write to nivie@wesleyan.edu.