The RAG Pipeline: Build It Step by Step

Five Stages

Every RAG system follows the same pipeline: ingest documents, split them into chunks, embed the chunks, store vectors with metadata, then retrieve and generate on each question.

Minimal Working Pipeline

pip install chromadb sentence-transformers

from sentence_transformers import SentenceTransformer
import chromadb

model = SentenceTransformer("all-MiniLM-L6-v2")
client = chromadb.PersistentClient(path="rag_notes")
notes = client.get_or_create_collection("notes")

notes.add(
    ids=["1", "2"],
    documents=[
        "RAG retrieves context before the model answers.",
        "Chunking splits documents into searchable pieces.",
    ],
)

question = "What is retrieved before answering?"
hits = notes.query(query_texts=[question], n_results=1)
print(hits["documents"][0][0])

Adding the Generation Step

context = hits["documents"][0][0]
prompt = f"Answer using only this note.

Note: {context}

Question: {question}"
# answer = call_your_llm(prompt)   # plug in any model or API you use

Why This Structure Wins

Storing chunks (not whole documents) makes retrieval precise, and metadata lets you filter by source, date or section.

Key Points

• Ingest, chunk, embed, store, retrieve, generate.
• Small real pipelines run with open-source libraries.
• Retrieved chunks become the model's context.
• Metadata keeps retrieval filterable.