Designing neural network architectures is challenging for learners who are new to machine learning, not only due to the abstract nature of data flow and tensor shape transformations across layers, but also the complexity of programming frameworks like PyTorch. Students often struggle to visualize how layer choices affect input and output dimensions, leading to errors and conceptual gaps. To address this, we present LayerStudio, an educational tool that allows learners to construct neural networks visually through a graph-based interface, which automatically generates corresponding, executable PyTorch code. The system validates network structure, ensures dimensional compatibility between layers, and produces both initialization and forward-pass methods. When errors occur, such as shape mismatches or invalid connections, learners receive informative feedback that reinforces architectural principles. By linking conceptual visualization with runnable implementations, our tool lowers the barrier to deep learning, supports active experimentation, and helps diverse learners develop an intuitive understanding of neural network construction.