In the world of machine learning, Keras's Model.fit has been a popular choice for training deep learning models. However, to truly understand the inner workings and gain fine-grained control over the training process, it is essential to go beyond the convenience of Model.fit and delve into building a custom training loop. In this talk, we will explore the implementation of a custom training loop using Python and TensorFlow, with a special focus on leveraging the powerful logging capabilities of Weights and Biases (wandb). Join us as we embark on a journey to unlock the potential of custom training loops and take our machine learning expertise to the next level.