The Brickcode Prediction project focuses on automating the classification of products into Global Product Classification (GPC) Brick Codes by building an intelligent machine learning pipeline. Using a blend of historical product data and real-time user-provided descriptions, the system leverages robust data preprocessing, feature engineering, and selection techniques to extract meaningful patterns and representations. A hybrid approach was used, combining string similarity methods with a Recurrent Neural Network (RNN)-based Bidirectional LSTM model. This cascade model first attempts rule-based predictions using string-matching for high-confidence matches, and then applies deep learning for more complex, ambiguous cases to enhance prediction accuracy and consistency.

To support scalability and performance, PySpark was employed throughout the pipeline for distributed processing of large datasets. This allowed the team to preprocess data, train models, and run predictions efficiently at scale. The overall solution ensures timely and accurate Brick Code predictions, which are crucial for maintaining standardized product classification across systems.