The paper introduces the Prime Video Page Composition framework (POCN), comprising Long Term Customer Propensity, Occasional Exploration, Contention Relevance, and Neighborhood Diversity, for optimal carousel positioning on the homepage. The framework addresses the challenge of balancing diverse customer and business considerations, including different offers (Prime/individual purchase/3P) and content types (Movie/TV/Sports). To achieve this, a Transformer model predicts long-term customer propensities for various offer/content types, integrating discounted impression UCB for occasional exploration. A novel neural network captures contention between carousels, and the framework combines long-term propensity, occasional exploration, and contention relevance linearly, with added neighborhood diversity. The proposed approach demonstrates a significant (4.6%) improvement in customer engagement metrics in A/B experiments.