How might low- and middle-income countries ensure embedded AI tech allows for alignment with local curriculum, assessment, and teacher needs?
How might low- and middle-income countries ensure embedded AI tech allows for alignment with local curriculum, assessment, and teacher needs?
EdTech Hub’s AI Observatory is exploring how education systems can create context-driven solutions – which means ensuring AI tools, models, or approaches genuinely meet the diverse needs of a place and the people within it.
This week, in Issue No. 28 of the #WaypointWednesday, we spotlight context-aware benchmarks, experts-in-the-loop, and deploying LLMs via WhatsApp.
Early signals
Context-aware benchmarks for evaluating AI systems
We’re seeing benchmarks and evaluations designed specifically to assess how well AI systems align with the local contexts in which they are used.
- Global – Common Sense: Common Sense AI risk assessments take into account the societal landscape in which the products will be used, asking questions like “Whose interests, desires, skills, experiences, and values might this product have simply assumed, rather than actually consulted?” (Common Sense)
- Latin America – CENIA’s cultural benchmark: Chile’s National Center for Artificial Intelligence (CENIA) developed a test to compare how well LLMs represent the knowledge of Latin American people, and revealed that while mainstream models easily identify Buenos Aires, they fail to recognise common regional foods or important local figures. (Nature, 2025)
Experts-in-the-loop to localise educational resources
We’re seeing efforts to combine the speed of AI with the expertise of educators to accelerate the creation and adaptation of learning resources for local needs.
- Benin, Cameroon, and the Democratic Republic of Congo – GPE KIX project: To address gaps in relevant STEM learning resources, Open Educational Resources were localised three times: first by the AI tools, then by local reviewers, and finally by the translation team. (GPE KIX, 2025)
- India – Chimple: At a recent EdTech Hub event, Srikanth Talapadi, Managing Director of Chimple, described how Chimple places current teachers at the center of content generation, providing them with AI workbenches that help generate contextually relevant content. (EdTech Hub, 2025)
Deploying Large Language Models (LLMs) via WhatsApp
Where access to computers and reliable internet is limited, we’re seeing LLMs delivered through lower-bandwidth channels like WhatsApp in an effort to lower costs and barriers to access.
- Ghana – Rori: Rori is an AI-powered maths tutor delivered through WhatsApp that has been shown to improve learning outcomes for students in Ghana. The intervention was specifically designed to work with basic mobile devices on low-bandwidth data networks. (Henkel, 2024)
- Sierra Leone – TheTeacher.AI: TheTeacher.AI is a WhatsApp chatbot that teachers preferred to web search because it can create localised content and uses far less data. Some teachers, however, struggled to use it effectively, highlighting the need for training. (Björkegren et al., 2025)
Reflections:
- When designing contextually, it is important to note that there is often not just one, but many cultures and languages within a particular context that are shaped by complex and overlaid histories, power relations, and socio-political forces.
- With increasingly hybrid identities, learners and teachers may experience cognitive dissonance and have to navigate multiple, sometimes conflicting, cultural and knowledge systems, so contextual design must cater to pluralistic worldviews to help negotiate these tensions.
Swipe for a quick take 👇🏽
We’d love to hear from you! What’s been shaping your thinking on AI? Drop your thoughts (and reading recommendations) in the comments. Explore more from EdTech Hub’s AI Observatory.
EdTech Hub’s AI Observatory is made possible with the support of the UK’s Foreign, Commonwealth and Development Office.
