Neural networks (NNs) are a fundamental component of AI-enhanced software systems, often built following model-driven engineering (MDE) / low-code approaches. Yet, while MDE tools support the modelling of most system components, designing and implementing NNs remains a manual, repetitive, and error-prone task that requires expertise in framework-specific APIs such as PyTorch or TensorFlow. In this paper, we address this gap by extending the BESSER web modelling editor to support NN diagram design and automatic code generation. Our tool relies on a formal NN metamodel, which provides a framework-agnostic representation of neural networks that is automatically converted to runnable PyTorch and TensorFlow code, covering both architecture definition and training and test processes. We validate our tool on a set of representative NN architectures and demonstrate correct code generation across both frameworks. By building on the BESSER platform, our work paves the way for treating NNs as modelled artifacts within a unified MDE workflow. Our tool is open-source, released under the MIT license, and accessible at https://editor.besser-pearl.org/. A video demonstration is available at https://youtu.be/DCb6qwnnlQc

@inproceedings{paperbot12566,
  title = {BESSER-NN: Visual Neural Network Design and Automated Code Generation with the BESSER Low-Code Platform},
  author = {Nadia Daoudi and Armen Sulejmani and Jordi Cabot},
  booktitle = {ASE},
  year = {2026},
  doi = {10.1145/3832783.3834595}
}