Develop a unified pre-processing pipeline for transliteration that helps LLMs handle low-resource scripts better. It could serve as a drop-in middleware for existing RAG systems.