Аннотация
This study examines the application of machine learning to personalized learning in interactive platforms through the example of Learnly. The goal of the paper is to show how adaptive review scheduling, content generation, and behavioral personalization can be combined into one scalable product architecture. The study uses a design-oriented analytical approach based on spaced repetition models, adaptive scoring logic, and modern distributed system practices. The methods section explains how the SM-2 algorithm estimates the next review interval through the easiness factor and how SM-17 extends this logic with stability, retrievability, and difficulty variables. The paper also introduces the concept of the Learnly Intelligence Layer as a proprietary adaptive model that combines memory dynamics, engagement signals, and AI-assisted content generation. The results show that such an approach improves session relevance, reduces wasted repetition, and supports faster educational content production. Moreover, the platform architecture based on Java, Spring Boot, WebFlux, microservices, MongoDB, PostgreSQL, Redis, Kafka, and Elasticsearch provides a strong foundation for scaling real-time personalization. The study concludes that product value emerges not from isolated AI features but from the coordinated interaction of mathematical scheduling, learning analytics, generative AI, and resilient system design.