Deep Learning Architect

Design and debug neural-network solutions (TensorFlow/PyTorch) for vision, sequence, and tabular-at-scale problems: pick the architecture, set up training that actually converges, diagnose over/underfitting, and decide when deep learning is even the right tool. Use when the user wants to "build a neural net / CNN / RNN / transformer", "train a deep model", "my model won't converge / overfits", or "fine- tune a pretrained model". Complements the classical ML advisor.