Sub-Saharan Africa faces persistently high student-to-teacher ratios. A single science teacher may be responsible for over a hundred students, making individualized feedback impossible. But sending a WhatsApp message? That's easy—WhatsApp penetration in West Africa is extremely high, used by nearly every student and parent. Adesua, developed by Boateng, Atompoya, and their team, is an AI science tutor built directly on WhatsApp.
How It Works
The technical architecture is not complicated: the backend integrates curriculum standards and a 33-year archive of national exam questions, while the frontend interacts with students through a WhatsApp Bot interface. A student sends a message like "explain photosynthesis," and the AI retrieves relevant textbook content to generate an answer. Students can also take scheduled or on-demand multiple-choice quizzes organized by topic or by year; after submitting, they immediately receive a score with explanations for each question.
Why Simplicity Is a Feature
The lack of technical complexity may actually be an advantage. In resource-constrained regions, Adesua requires no app store download, no account registration, and no additional data plan—WhatsApp traffic is free, and messaging is an interaction pattern students already know. This means zero deployment friction.
Feasibility Study Results
The feasibility experiment ran for 6 months with 56 active users (including students and parents). The helpfulness rating for AI-generated answers was 93.75%—but based on only 16 ratings, an extremely small sample. The figure is preliminary and fragile, though it provides a starting point for larger-scale research.
Design Philosophy: Meet Students Where They Are
An interesting angle: Adesua shares a design philosophy with Eskwai (another paper from the same group at the same venue)—build AI education on the communication infrastructure students already have, rather than requiring them to adapt to a new platform. A WhatsApp bot is more likely to be genuinely used than a standalone education app, simply because it doesn't demand a change in user behavior.
Open Questions
- Is 56 active users over 6 months a high or low retention rate in a real-world setting?
- The 93.75% helpfulness rating rests on only 16 rating samples—limited statistical significance.
- Has answer accuracy been rigorously evaluated? Could the AI deliver incorrect scientific knowledge to West African secondary students?
- Have the exam question bank answers been manually verified?
References
1. Boateng, G., Atompoya, E., Badu, P., et al. (2026). *Adesua: Development and Feasibility Study of an AI WhatsApp Bot for Science Learning in West Africa*. arXiv:2605.15376 [cs.CL]. 2. Boateng, G., et al. (2024). *Kwame for Science: A Learning and Assessment Platform for Science Education in Africa*. AIED. 3. UNESCO. (2023). *Technology in Education: A Tool on Whose Terms?* Global Education Monitoring Report.