Based on a distributed microservices architecture, this paper develops an intelligent recommendation system for fashion retail. In order to achieve millisecond recommendation response, this system combines deep learning models with a real-time computing framework. The recommendation accuracy and user shopping conversion rate are significantly improved by optimising feature engineering and constructing a product knowledge graph. Multi-tier fault-tolerance mechanisms help maintain stable performance and high availability as well. For the fashion retail industry, this provides a reliable technical solution for personalised recommendations. In practice, the system is an effective improvement to the user’s shopping experience and the platform’s operating efficiency.