Design and Technical Validation of the “AI Edu” Learning Platform with Grounded RAG and Interactive Virtual Laboratories

This study presents the design and technical validation of AI Edu, an educational platform combining course delivery, retrieval-augmented generation (RAG), progress records, and interactive virtual laboratories. The architecture separates the portal, domain backend, assistant gateway, and browser computation. A multilingual RAG experiment used GPT-6 Luna, 144 synthetic resources, and 108 questions. Hit Rate@6 was 100.0% and mean Precision@6 was 20.4%; all 72 answerable responses contained the expected numeric value, and all 36 unsupported questions produced the prescribed refusal. Current-source checks passed 185 backend tests, 142 frontend tests, 69 connected integration criteria, and 50 gateway criteria. Separate PostgreSQL and storage checks verified scope restrictions and record maintenance. The revised gateway implements course-only instructions, localized refusals, cancellation, and course/version consistency. The browser laboratory combines fixed convolutional filters with a trained dense classifier. It correctly classified 197 of 200 held-out synthetic glyphs (macro-F1: 0.9850) and 5773 of 10,000 Modified National Institute of Standards and Technology (MNIST) test images without retraining (macro-F1: 0.5741). Numerical replay verified arrays supplied to visualization. The results connect course-grounded assistance, verified progress, and executable learning objects within a common educational design.