Обещал пример RAG на LLamaIndex, о трех частях:
1) bm25 индекс

from llama_index.core import SimpleDirectoryReader
from llama_index.core.node_parser import SentenceSplitter
from llama_index.retrievers.bm25 import BM25Retriever

import Stemmer

documents = SimpleDirectoryReader("./data").load_data()
splitter = SentenceSplitter(chunk_size=512, chunk_overlap=256)
nodes = splitter.get_nodes_from_documents(documents)

bm25_retriever = BM25Retriever.from_defaults(
    nodes=nodes,
    similarity_top_k=2,
    stemmer=Stemmer.Stemmer("russian"),
    language="russian",
)
bm25_retriever.persist("./bm25_retriever")