Page 637 - AI for Good Innovate for Impact
P. 637

AI for Good Innovate for Impact



                   Use case -4: AI-Powered Offline Learning for Higher Education in

               Crisis Zones                                                                                         4.7: Education














               Country:               Palestine

               Organization:          Al-Aqsa University, Gaza

               Contact Person(s):

                    Shadia Y. M. Baroud (sy.baroud@ alaqsa .edu .ps)
                    Abdelrafe Elzamly (abd _elzamly@ alaqsa .edu .ps) - Head of Department of AI and
                    Computer Science


               1      Use Case Summary Table

                Category          Education

                The problem to be  War-induced infrastructure damage disrupts higher education; students
                addressed         lack reliable power, internet and personal computers.

                Key aspects of the  Education (Education technology, Accessibility, Resilient learning, Offline AI)
                solution          •  Agent-based AI (Multi-agent Systems), Wi Fi mesh, knowledge base
                                  •  Portable Raspberry Pi “AI servers” host content & tutors offline.
                                  •  Mobile-first access via local Wi-Fi
                                  •  AI-driven content recommendation & adaptive tutoring.
                                  •  • On-device auto-grading and analytics.

                Technology        AI agents, Offline AI, Edge computing, NLP tutor, Low-power hardware,
                keywords          Adaptive learning, Raspberry Pi, Solar-powered servers.
                Data availability  Mixed: Public OER datasets; Local student-interaction data (private, stored
                                  on-device).
                Metadata (type of  Text, code snippets, speech (TTS/STT), images/diagrams, assessment logs.
                data)
                Model training  Lightweight transformer-based NLP models (e.g., DistilBERT) fine-tuned
                and fine-tuning   offline; Matrix-factorisation for recommendations; on-device quantization
                                  (TensorFlow Lite).

                Code Reposito- University Moodle, Internal GitLab (private); plan to open-source core
                ries              modules after security review.













                                                                                                    601
   632   633   634   635   636   637   638   639   640   641   642