8.1.1 RAG Roadmap: Documents, Retrieval, Answers

RAG solves a practical problem: the model does not know every fresh, private, or source-specific fact, so the application must retrieve evidence before asking the model to answer.

See the RAG Pipeline First

The core loop is: load documents, split chunks, add metadata, embed, retrieve, rerank, assemble context, answer, cite sources, and evaluate.

Run a Tiny Retrieval Check

This is not a vector database yet. It is a tiny offline version of the retrieval habit: score chunks, print sources, and verify whether the evidence matches the question.

chunks = [
    {"source": "rag.md", "text": "RAG retrieves source chunks before the model answers."},
    {"source": "eval.md", "text": "Citations let users verify whether an answer is grounded."},
    {"source": "deploy.md", "text": "Deployment exposes the model through a stable API."},
]

query = "why do RAG answers need citations"
query_terms = set(query.lower().split())

def score(chunk):
    words = set(chunk["text"].lower().replace(".", "").split())
    return len(query_terms & words)

for chunk in sorted(chunks, key=score, reverse=True)[:2]:
    print(chunk["source"], score(chunk))

Expected output:

rag.md 2
eval.md 1

If the top source is unrelated, do not tune the final prompt first. Check document parsing, chunking, metadata, and retrieval coverage.

Learn in This Order

Step	Read	Practice Output
1	RAG basics	Draw the question → evidence → answer loop
2	Document processing	Produce chunks with source and metadata
3	Vector databases	Explain embedding, vector record, and similarity search
4	Retrieval strategies	Compare keyword, vector, hybrid, filter, and rerank
5	Optimization and advanced RAG	Debug poor recall, poor ranking, and weak context
6	RAG evaluation	Test answer correctness, citation support, and no-answer behavior

Evidence to Keep

Keep this page’s proof of learning as a small evidence card:

Query

one user question or test case

Retrieved Chunks

chunk ids, scores, and source titles

Answer

final response with citation or source note

Failure Check

missing evidence, wrong chunk, stale doc, or unsupported claim

Next Action

chunking, embedding, reranking, prompt, or eval change

Pass Check

You pass this chapter when you can build a minimal knowledge-base Q&A loop that prints retrieved chunks, answer text, and source citations for at least 10 fixed questions.

The exit mini project is a course knowledge-base assistant with 3 to 5 Markdown documents, top-k retrieval output, source display, and a simple evaluation table.

Check reasoning and explanation

1. A passing answer traces the full path from query to chunks, retrieval scores, cited evidence, answer, and fallback behavior.
2. The evidence should include retrieved passages, source metadata, a cited answer, and at least one empty-retrieval or wrong-retrieval case.
3. A good self-check explains whether a failure came from chunking, retrieval, ranking, prompt assembly, missing sources, or unsupported generation.