Agile software development is characterized by short iterative cycles known as sprints which aims to deliver high-quality software efficiently. Although multi-release reliability models exist, there is limited research on modeling and predicting reliability within a multi-sprint context. This paper presents a neural network-based model for predicting software reliability across multiple sprints. Using historical sprint data from similar projects, the model captures patterns to predict faults in ongoing sprints. The proposed approach outperforms traditional models, especially in early testing phases, by using previous sprint data. The model is trained and tested on the JIRA dataset, demonstrating superior performance compared to existing models, such as the Rawat and Saraf models, across various metrics, including MSE, Adjusted R-squared, and AIC. Results demonstrate the model’s applicability and accuracy using real-world software failure data from sprint-based development projects.