As part of a competitive capstone project at UT, I developed a neural network to predict which online shopping platform (A, B, or C) a customer is most likely to choose based on platform- and customer-specific features.

Key Highlights:

- Transformed 7,000+ customer choice records into a 3×4 matrix format to model each platform's features invariantly.

- Built and trained a custom neural network architecture that satisfies permutation invariance—ensuring that predictions remain consistent regardless of the order of platform inputs.

- Conducted iterative validation using accuracy metrics and loss curves to optimize generalization on unseen test data.

- Delivered a fully functional and modular pipeline in PyTorch, including preprocessing, training, validation, model saving, and a standalone test() function for probabilistic inference.

This project sharpened my skills in deep learning, model architecture design, and real-world decision modeling under uncertainty—especially with noisy and potentially irrational consumer behavior.