Synthetic Data

We develop techniques for generating high-quality synthetic training data, particularly focused on document and form understanding. Our methods improve model robustness and reduce reliance on expensive manual annotation.