|
PresentationsAn Approach to Creating Intelligent Agents to Support and Manage LMS Moodle Using a Large Language Model, Model Context Protocol, RAG, and LangGraphFederal State Budgetary Educational Institution of Higher Education Dubna University, Russian Federation, 141982, Moscow region, Dubna, Universitetskaya str., 19, Tel. 8(988)3403705, 8(929)1063027, medvedeva.j.j@uni-dubna.ru In the context of scaling the electronic information and educational environment of Dubna University and the continuous growth in the number of LMS Moodle users, the flow of repetitive user requests is increasing, creating an excessive load on the technical support service. Traditional automation methods often require the complex development of unique connectors, complicating their implementation. To relieve specialists and create an intelligent agent capable of handling the first line of consultations, it was decided to use the new open standard Model Context Protocol (MCP). This approach allows for unifying data exchange and integrating disparate university tools without rigid coupling to their software interfaces. The proposed technical solution integrates into Moodle and acts as a link between a fine-tuned large language model deployed in the Yandex Cloud infrastructure and the user. A key role in the system is played by the MCP server, which provides the agent with a set of executable tools developed at the university by ISAU staff for intelligent processing, filtering, and routing of user requests. The LangGraph framework is used to manage dialogue logic: during request processing, the model analyzes the context and independently determines the problem-solving strategy. When receiving mass typical requests regarding course navigation or working in the system, the agent uses a developed RAG tool to generate an instant response, allowing the ticket to be closed in real time. The tool for creating an official ticket in the technical support system is used only in situations where the request is classified as atypical or requires administrative privileges. This allows a significant number of requests to be fulfilled without involving technical specialists, directing only verified incidents requiring expert intervention. The modularity of the MCP-based architecture facilitates system scalability: new functions are added on the server side without making changes to the agent core code or LMS Moodle itself.
|
