RNA’s functional diversity arises from its rich three-dimensional structural landscape, yet designing RNA sequences that adopt a target 3D conformation (inverse folding) remains a fundamental challenge. I present a geometric deep learning framework that enables end-to-end RNA inverse folding by integrating Geometric Vector Perceptron (GVP) layers into a Transformer architecture, thereby explicitly encoding the spatial and directional features of RNA structures. A curated dataset of experimentally solved RNA 3D structures was constructed from the BGSU RNA database through rigorous filtering and deduplication. Model performance was evaluated using sequence recovery rate and TM-score to assess sequence accuracy and structural fidelity, respectively. Across standard benchmarks and RNA-Puzzles, the proposed method achieves state-of-the-art performance, improving the recovery rate by 85% and the TM-Score by 40% compared to existing approaches. Masked family-level validation using Rfam annotations demonstrates strong generalization to unseen RNA families, with a 132% improvement in generalization performance. Furthermore, inverse-folded sequences refolded using AlphaFold3 closely recapitulate native structures, underscoring the importance of explicit geometric representations in RNA design. This work establishes a “3D-aware” paradigm for RNA inverse folding, enabling the rapid generation of structurally valid RNA sequences in hours rather than months. By combining geometric learning with sequence modeling, the framework offers a scalable foundation for accelerating mRNA vaccine development, RNA-based therapeutics, and synthetic biology, and positions geometric deep learning as a key technology for next-generation biomolecular design.