Architecture And Experimental Evaluation Of A Cross-Platform Mobile Application For Adaptive Learning Using Large Language Models

Abstract

The paper addresses the design and experimental evaluation of the architecture of a cross-platform mobile application for adaptive foreign language learning using large language models. An architectural approach based on a dedicated AI Integration Layer is proposed, enabling the separation of business logic and improving the reliability of interaction with external AI services. The system implements adaptive content generation considering individual user characteristics. Special attention is given to performance optimization through the implementation of a multi-level caching mechanism, which reduced AI service usage costs by 74 % and decreased response latency. An experimental evaluation of system performance demonstrated stable operation under a load of up to 100 concurrent users with an average response time of 280-500 ms, and identified a degradation threshold at 160-170 users. The obtained results confirm the effectiveness of the proposed approach and its applicability for developing modern mobile educational systems based on large language models.