Gender bias in legal documents is often subtle and depends on context rather than easily identifiable keywords, making manual expert review costly and difficult to scale. This work proposes a processing pipeline that combines natural-language processing, text embeddings, binary classification, and large language models to support the automatic detection and explanation of gender bias in judicial texts. An initial evaluation on judicial rulings examines both classification performance and the usefulness of LLM-generated explanations. The results are promising in terms of precision and recall and suggest that combining conventional classification with language-model explanations can help analysts identify potentially biased passages while retaining contextual information needed for expert assessment.