How does RAG handle multi-lingual knowledge bases?

Multilingual RAG serves users in their preferred language while retrieving from documents in any language in the knowledge base. Approaches include multilingual embedding models like multilingual-e5-large and mGTE that map different languages into a shared embedding space enabling cross-lingual retrieval. A query in English can retrieve relevant French documents and vice versa. The LLM then generates the response in the user language regardless of the source document language. Alternatively, translate documents to a common language during indexing for simpler retrieval but at the cost of translation artifacts. For best results, combine multilingual embeddings with language-aware metadata filtering to prefer same-language matches when available. Test cross-lingual retrieval quality for each language pair in your use case since performance varies. Consider maintaining parallel indexes for high-priority languages to optimize retrieval quality within each language.