In contemporary higher technical education, academic institutions affiliated with large engineering universities face significant hurdles in monitoring student learning trajectories, diagnosing root conceptual deficiencies, and providing timely faculty interventions. Traditional examination systems rely predominantly on post-hoc summative grade sheets that report aggregate percentages without uncovering why a student struggles. Furthermore, standard Learning Management Systems (LMS) and black-box predictive models provide ungrounded score predictions, lack pedagogical explainability, suffer from generative AI hallucinations, and neglect strict multi-tenant institutional privacy boundaries. To overcome these fundamental challenges, this paper presents EDU-VISION, an end-to-end, explainable academic intelligence and diagnostic learning gap framework designed for engineering curricula. EDU-VISION integrates a domain-constrained Retrieval-Augmented Generation (RAG) pipeline, deterministic student performance profiling, an explainable non-judgmental learner classification model, a prioritized six-category diagnostic learning gap detection engine, and an assignment-scoped faculty intervention lifecycle.

By grounding exam generation and evaluation in verified university marking schemes, the system eliminates metric fabrication. Furthermore, deterministic algorithms classify learner progression across four evidence-backed tiers, diagnose specific cognitive gaps (Concept, Answer Writing, Scheme Alignment, Application, Revision, and Consistency), and recommend verified pedagogical resources. Faculty members receive assignment-scoped intelligence dashboards with non-causal longitudinal tracking, protected by enterprise-grade distributed rate limiting and multi-college boundary isolation.

Keywords— Educational Intelligence, Retrieval-Augmented Generation (RAG), Learning Gap Detection, Explainable AI, Faculty Intervention, Academic Analytics, Multi-Tenant Architecture.