What Is RAG?

1. The Problem RAG Solves

When you ask an AI model a question, it answers from memory — everything it learned during training. That training data has a cutoff date, doesn't include your private documents, and is sometimes simply wrong (this is called "hallucination").

Imagine asking an AI: "What is the attendance policy in our college handbook?" The AI has never seen your college handbook. So it will either make something up or admit it doesn't know.

RAG solves this by giving the AI access to a specific set of documents at the moment it answers. Instead of relying purely on memory, it first searches your documents, retrieves the most relevant sections, and then uses those sections to generate an accurate answer.

2. The RAG Pipeline — How It Works

RAG has two phases: an indexing phase (done once, when you add documents) and a retrieval phase (done every time someone asks a question).

3. Build a Simple RAG System in Python

We'll use ChromaDB (free, local vector database) and Google Gemini for generation. No paid services needed.

Install Libraries

pip install chromadb google-generativeai pypdf

Full RAG Script

import os
import chromadb
import google.generativeai as genai
from pypdf import PdfReader

# ── Config ────────────────────────────────────────
genai.configure(api_key=os.environ["GEMINI_API_KEY"])
model = genai.GenerativeModel("gemini-1.5-flash")
chroma = chromadb.Client()
collection = chroma.get_or_create_collection("my_docs")

# ── Step 1: Load and chunk a PDF ──────────────────
def load_pdf(path, chunk_size=400):
    reader = PdfReader(path)
    text = " ".join(page.extract_text() for page in reader.pages)
    words = text.split()
    chunks = []
    for i in range(0, len(words), chunk_size):
        chunk = " ".join(words[i:i+chunk_size])
        chunks.append(chunk)
    return chunks

# ── Step 2: Index the document ────────────────────
def index_document(pdf_path):
    chunks = load_pdf(pdf_path)
    ids = [f"chunk_{i}" for i in range(len(chunks))]
    collection.add(documents=chunks, ids=ids)
    print(f"Indexed {len(chunks)} chunks from {pdf_path}")

# ── Step 3: Answer a question using RAG ───────────
def ask(question, top_k=3):
    results = collection.query(query_texts=[question], n_results=top_k)
    context = "\n\n".join(results["documents"][0])
    prompt = f"""Use the following context to answer the question.
If the context doesn't contain the answer, say "Not found in document."

Context:
{context}

Question: {question}
Answer:"""
    response = model.generate_content(prompt)
    return response.text

# ── Run ───────────────────────────────────────────
index_document("your_document.pdf")

while True:
    q = input("\nAsk a question (or 'quit'): ").strip()
    if q.lower() == "quit":
        break
    print("\n" + ask(q))

Replace your_document.pdf with any PDF — a textbook, a datasheet, your college syllabus, lab manual, or project report.

4. Popular RAG Tools and Frameworks

• LangChain — the most popular RAG framework; handles chunking, embedding, retrieval, and generation with minimal code
• LlamaIndex — specialised for document Q&A; easier to use for beginners than LangChain
• ChromaDB — free, local vector database; runs on your laptop, no account needed
• FAISS — Meta's vector similarity library; ultra-fast for large document sets
• Pinecone — cloud vector database; needed for production apps with many users

5. Real Project Ideas Using RAG

• Chat with your college placement brochure to get company-specific info
• Q&A bot for a manufacturer's product datasheet
• Study assistant trained on lecture notes and past question papers
• Customer support bot for a local business
• Legal document assistant (e.g., rent agreement explainer)