JASP users often need extra help when learning how to use JASP or understand statistical procedures. If they ask general LLMs for help, the answers may be incorrect or hallucinated. If they search for official help, the documentation is spread across many different sources, making it hard to find the right information quickly. To solve these problems, this thesis develops a multimodal retrieval-augmented generation (RAG) protocol that can reliably retrieve the correct JASP documentation to provide grounded answers. The system consolidates GitHub markdown files, PDF manuals, and YouTube tutorials into a unified knowledge base. Retrieval is carried out through an integrated pipeline combining keyword searching, semantic searching, fusion, and reranking methods. An interactive Streamlit interface allows users to inspect retrieved sources directly. Retrieval performance is evaluated with a curated test set spanning multiple difficulty levels, enabling systematic comparison of different retrieval strategies and analysis of their effectiveness on simple, complex, and ambiguous queries.

Overall, this thesis delivers the first domain-specific, open-source RAG framework designed for JASP. It provides a scalable retrieval architecture, a complete multimodal pipeline, and a systematic evaluation of retrieval strategies. Together, these contributions form a strong foundation for future intelligent support tools in statistical software.