Project 02: Neural Network Classifier

Objective

Build a small image or text classifier with a reproducible training pipeline, experiment tracking, evaluation, export, and inference API.

Learning Outcomes

• Prepare data for deep learning.
• Train neural networks with a modern framework.
• Track experiments.
• Diagnose overfitting and underfitting.
• Export and serve a model.

Tasks

• Choose an image or text classification dataset.
• Create train, validation, and test splits.
• Build a baseline neural network.
• Add regularization.
• Track hyperparameters and metrics.
• Export the model.
• Build an inference API.
• Add Docker deployment.
• Write an experiment report.

Deliverables

• Training script.
• Evaluation script.
• Inference API.
• Model artifact export instructions.
• Experiment report.
• Dockerfile.

Tips

• Verify the model can overfit a tiny batch before full training.
• Plot training and validation loss.
• Keep preprocessing identical between training and inference.
• Save model version and dataset version together.