Offline Retrieval Augmented Artificial Intelligence for Academic Support in Low Connectivity Namibian Universities
Offline academic support requires evidence that is usable, institutionally approved and appropriate to the date of a question. This technical note specifies a retrieval-augmented architecture for low-connectivity Namibian university settings, integrating document validity, extraction quality, citation checks and controlled abstention. A deterministic verification experiment examines 32 synthetic metadata fixtures across four authorization policies. The full gate authorizes the two eligible fixtures and blocks all 30 ineligible fixtures; these counts establish rule conformance, not language-model accuracy. Analytical weight-storage calculations clarify the resource implications of model size and quantization. Comparisons with recent studies distinguish temporal retrieval, context sufficiency and compression from the proposed governance integration. A prospective factorial evaluation is specified for lecturer-verified questions, factual and citation accuracy, latency, memory and energy. The contribution is an auditable design and limited computational verification; device feasibility and educational effectiveness remain to be established.
Authors
- Philipus Nghifikepunye Nangolo
- Chiyesu Stefanus
- Kaarina Ndeapo Nakale
Institutions
- International University of Management (NA)
Publication Details
- Journal
- Engineering and Technology Journal
- Published
- 2026-11-14
- DOI
- https://doi.org/10.47191/etj/v11i11.05
- Primary Topic
- Artificial Intelligence in Education
- Type
- article
- Field-Weighted Citation Impact
- 0.00