Design patterns from SAGE-RAI for making transparency a core part of RAG systems, not an optional extra.
Retrieval-augmented generation (RAG) is increasingly used in web-based education — pulling trusted materials into generative answers. Transparency is too often treated as an ethical add-on. This short paper argues it should be architectural.
Drawing on the design and deployment of SAGE-RAI, an advanced multi-purpose RAG system for learning, we set out design patterns for transparent RAG. Evaluation combined quantitative ratings (n=26, mean 4.62/5) with qualitative interviews (n=4). Learners were highly satisfied (92.3% rated 4–5 stars), while also highlighting tensions between AI assistance and learning independence.
As RAG systems mediate more of how people reach web-based knowledge, the paper makes the case that transparency has both pedagogical and ethical work to do — and needs to be built in from the start.
Kwarteng, J.; Third, A.; Mikroyannidis, A.; Tarrant, D. and Domingue, J. (2026). SAGE-RAI: Design Patterns for Transparent RAG Systems. In: WWW ’26: Proceedings of the ACM on Web Conference 2026, pp. 8641–8644.
Download the published version (PDF) · View on ACM / DOI · SAGE-RAI publications
See how ODI learning connects data, AI and trustworthy practice on our courses.