Mastering RAG

Fundamentals

RAG grounds LLM responses in retrieved data, reducing hallucinations and improving accuracy. Key components: embedding model, vector store, and LLM.

LangChain & Vector Databases

pip install langchain chromadb sentence-transformers

Chunking and Indexing

from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain_community.embeddings import HuggingFaceEmbeddings

splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
chunks = splitter.split_text(document)

embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
vectorstore = Chroma.from_texts(chunks, embeddings)

Complete RAG Pipeline

from langchain.chains import RetrievalQA

qa = RetrievalQA.from_chain_type(
    llm=llm,
    retriever=vectorstore.as_retriever(search_kwargs={"k": 3})
)
result = qa.invoke("What is the capital of France?")
print(result)

RAG Evaluations

• Faithfulness — Does the answer match the retrieved context?
• Relevance — Is the retrieved context relevant to the query?
• LLM-as-Judge — Use an LLM to rate answer quality

Advanced RAG Techniques

• Pre-processing — Clean documents before chunking
• Re-ranking — Re-rank retrieved chunks with a cross-encoder
• Query Expansion — Generate multiple queries from one question
• GraphRAG — Build a knowledge graph from documents for better retrieval