I have a text-classification dataset where a few classes dominate the rest, and I want to correct that skew with a Generative Adversarial Network. The objective is straightforward: generate convincing synthetic samples for the minority classes so the final corpus is evenly distributed and ready for model training. You’ll start from the raw, imbalanced text I provide, build or adapt a GAN architecture suited to natural-language generation, and iterate until each class reaches parity without sacrificing linguistic quality. I’m open to your preferred framework—PyTorch, TensorFlow, or a lightweight alternative—as long as the code is clean, reproducible, and clearly documented. When we’re done, I expect: • Python code for data preprocessing, GAN training, and synthetic text generation. • A report (not long, just clear) that shows class counts before and after, explains the architecture you chose, and includes evaluation metrics or sample outputs that demonstrate realism. • Instructions for me to rerun or extend the process on new data. If this sounds like the kind of hands-on, results-focused project you enjoy, let’s get started.