This paper explores the application of Retrieval-Augmented Generation (RAG) in multilingual settings, focusing on building a robust baseline for future research. The authors investigate the components and adjustments needed to create an effective multilingual RAG pipeline, which can handle user queries and datastore in 13 languages. Key findings include the need for task-specific prompt engineering to enable generation in user languages, adjustments to evaluation metrics to account for variations in named entity spelling, and the importance of strong multilingual LLMs for accurate and fluent generation. The study also highlights limitations such as code-switching in non-Latin scripts, fluency errors, and irrelevant retrieval. The authors release their mRAG baseline pipeline and provide insights into future research directions, emphasizing the need for stronger multilingual LLMs, LLM-based evaluation, and multi-domain multilingual retrieval.This paper explores the application of Retrieval-Augmented Generation (RAG) in multilingual settings, focusing on building a robust baseline for future research. The authors investigate the components and adjustments needed to create an effective multilingual RAG pipeline, which can handle user queries and datastore in 13 languages. Key findings include the need for task-specific prompt engineering to enable generation in user languages, adjustments to evaluation metrics to account for variations in named entity spelling, and the importance of strong multilingual LLMs for accurate and fluent generation. The study also highlights limitations such as code-switching in non-Latin scripts, fluency errors, and irrelevant retrieval. The authors release their mRAG baseline pipeline and provide insights into future research directions, emphasizing the need for stronger multilingual LLMs, LLM-based evaluation, and multi-domain multilingual retrieval.