Wikipedia[en] is a RAG index built using the '20220301.en' subset of HuggingFace dataset wikipedia. It contains ~6M cleaned articles in English. The RAG is built using the BAAI/bge-large-en-v1.5 model. We embedded full articles rather than just titles, coupled with summarization of long chunks to enhance retrieval performance significantly. The index also enables filtering retrieval context by filling a “query” input, simply some keywords. Then, along with the vectors you sent, the keywords will be used for extracting the exact passage from the context for cutting noise and providing only the most accurate part you need via the bm25 method.