Wubba learns layout-invariant embeddings from raw HTML using contrastive learning. Convert any HTML document into a fixed-size vector for similarity search, clustering, or classification.