Scholarlink

The integration of artificial intelligence (AI) into digital education is shifting from generic platforms to domain-specific, data-driven solutions. This paper introduces the Electronic Professor: An Intelligent Avatar for Learning Electricity, an AI-based framework developed during a research internship at the Faculty of Sciences Semlalia. The system supports students in mastering electrical science concepts by combining retrieval-augmented generation (RAG) with fine-tuning of open-source language models (LLMs). The approach relies on a heterogeneous educational corpus—textbooks, lecture notes, datasets, web content, and multimedia—processed into a FAISS-indexed knowledge base. A prototype using the Flan-T5-small model enables efficient retrieval and pedagogical question answering across more than 60 academic files. Early experiments show that the system enhances access to structured knowledge, reduces search time, and offers interactive features like quiz generation and chapter summarization. Ongoing work explores fine-tuning with larger models (Mistral, Phi-2). Results suggest that AI can foster engagement, personalization, and self-directed learning while complementing, not replacing, the educator’s role. This study contributes a scalable, open-source framework for intelligent tutoring systems, bridging raw heterogeneous resources with personalized digital learning.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Digital Education in Electricity Through an AI Avatar with RAG and Fine-Tuned LLMs

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Abdelali El Gourari,
  • Abir Zennir,
  • Mustapha Raoufi,
  • Mohammed Skouri

摘要

The integration of artificial intelligence (AI) into digital education is shifting from generic platforms to domain-specific, data-driven solutions. This paper introduces the Electronic Professor: An Intelligent Avatar for Learning Electricity, an AI-based framework developed during a research internship at the Faculty of Sciences Semlalia. The system supports students in mastering electrical science concepts by combining retrieval-augmented generation (RAG) with fine-tuning of open-source language models (LLMs). The approach relies on a heterogeneous educational corpus—textbooks, lecture notes, datasets, web content, and multimedia—processed into a FAISS-indexed knowledge base. A prototype using the Flan-T5-small model enables efficient retrieval and pedagogical question answering across more than 60 academic files. Early experiments show that the system enhances access to structured knowledge, reduces search time, and offers interactive features like quiz generation and chapter summarization. Ongoing work explores fine-tuning with larger models (Mistral, Phi-2). Results suggest that AI can foster engagement, personalization, and self-directed learning while complementing, not replacing, the educator’s role. This study contributes a scalable, open-source framework for intelligent tutoring systems, bridging raw heterogeneous resources with personalized digital learning.