With the growing significance of RNA in biotechnology, RNA design is becoming an essential part of drug discovery. Breakthroughs in deep learning have advanced our ability to address the 'folding problem' to predict the secondary and tertiary structure of RNA given sequence. However, also critical to RNA design is the 'inverse folding problem' where the RNA sequence is optimized to match a desired tertiary (and secondary) structure. In this work, we propose 3DRNA - a deep neural network model to automate the design of sequences on a fixed RNA backbone. By learning the spatial relationship of atoms in a residue's local environment, the model predicts the corresponding RNA base and its chi angle. Our preliminary results suggest the model achieves a 52.2\% overall sequence recovery, with 61.1\% for RNA-only structures and 49.6\% for RNA-complex structures.