Poetry generation has a huge advancement with the help of artificial intelligence by balancing creativity, thematic consistency, and structural adherence. While the prior advancements with the use of neural language models were promising, but they often lacked dynamic user interaction and refined outputs. In this work, an innovative Haiku generation framework is proposed, which integrates reinforcement learning and dynamic user feedback into Google’s Gemini 1.5 model. The proposed model is designed to enhance the quality and personalization features of the poetry generated by iteratively refining the outputs based on user-level provided feedback. The proposed framework incorporates four key stages, namely input validation and preprocessing, optimized prompt generation, poetic structure enforcement, and iterative refinement loops. The method achieves high structural fidelity (5-7-5 Haiku format) and also promises emotional resonance and semantic alignment with user-defined themes. The proposed method has been found to be outperforming the existing methods, as validated by quantitative comparisons and user evaluations. The approach advances AI’s creative potential and also redefines the process of interactive poetry generation by making the poems artistically meaningful and thematically rich.