RAG-Tutor: Generative AI Tutor for Self-Directed Learning in MOOCs

RAG-Tutor is an applied research project funded by the International Association for Smart Learning Environments (IASLE) and NetDragon Websoft, a Chinese company specialised in educational platforms. The project introduces a Retrieval-Augmented Generation (RAG) system designed to act as a tutor in Massive Open Online Courses (MOOCs), offering real-time interaction with students and supporting self-directed learning (SDL) and dialogue-based learning.

Addressing the Limitations of MOOCs

MOOCs face some limitations, such as limited interaction with instructors, and insufficient personalized feedback. RAG-Tutor tackles these issues by integrating information retrieval with large language models (LLMs) in order to provide answers. It uses this information to generate based in course materials and providing learners with relevant and accurate support. Its architecture is based on three key principles:

  • Privacy: All interactions and course data must remain private.
  • Scalability: Easily deployed across multiple courses and subjects.
  • Contextual Support: Answers generated from course resources, ensuring they are relevant and contextualized for the student.

How Does RAG-Tutor Work?

RAG-Tutor operates through three main stages. First, the system performs information retrieval using semantic search techniques. Then, it searches a database composed of relevant documents, such as manuals, articles, or study materials. To do this, it converts texts into numerical vectors that capture their meaning, enabling the system to identify and extract the relevant fragments for each query.

Second, once the appropriate fragments are retrieved, a generative language model—such as GPT-3.5 or GPT-4—uses this information to generate coherent and contextualized responses. This model is fine-tuned to ensure that the answers are accurate and aligned with the retrieved content.

Finally, RAG-Tutor personalizes and adapts its responses based on the student’s profile and progress. It uses Knowledge Tracing techniques to adjust both learning recommendations and generated responses, thereby enhancing the relevance and effectiveness of the educational process.

Designed for MOOCs

RAG-Tutor is designed as an adaptable framework that institutions can integrate into their MOOCs or document query systems. By supporting self-directed and dialogue-based learning, it contributes to more effective and engaging online education.

This project is implemented by the Institute for Research, Innovation, and Educational Technology (UNIR iTED), which is responsible for the design, evaluation, and development. In August 2025, during the Global Smart Education Conference 2025 held in Beijing, China, the RAG-Tutor project was awarded the Educational Innovation Award. This international recognition highlights the impact and relevance of RAG-Tutor in the field of smart digital education.

Funding type: Private

NetDragon

Timespan: 2025