M7-C — RAG with NotebookLM

How to read this note

NotebookLM is the easiest way to feel RAG working. Upload PDFs → ask questions → AI answers from your files only. No coding.

🎧 What is NotebookLM?

NotebookLM is Google's document chat tool.

Behind the scenes it runs the RAG pipeline from M7-B — How RAG Works Behind the Scenes:

You upload sources  →  Google indexes them  →  You chat  →  Answers cite your sources

Extra features:

Audio Overview — AI hosts discuss your notes like a podcast
Study guides / FAQs — auto-generated from uploads
Slide deck — draft presentation from your material

🚀 Setup (5 minutes)

Go to notebooklm.google.com
Sign in with Google account
Click New Notebook
Name it (example: Module 7 RAG Practice)
Click Add source → upload PDF, Google Doc, or paste text

Classroom exercise

Upload all your Module 1–3 Obsidian notes exported as PDF. Ask: "Explain the intelligence stack in simple words." Check if citations point to your files.

💬 How to chat well with NotebookLM

Bad question	Better question
"Tell me about AI"	"According to my notes, what is the difference between ML and LLM?"
"Everything about Module 3"	"List the 3 prompt components from my uploaded file"
"Is this true?"	"Quote the line from my document about hallucination"

NotebookLM works best when you point it at your material.

🔍 What happens in the backend (while you wait)

Upload click
    ↓
Google splits your file into chunks
    ↓
Each chunk → embedding → stored in Google's index
    ↓
You type question → similar chunks retrieved
    ↓
Gemini model reads chunks + writes answer
    ↓
Citations link back to source paragraphs

You never see steps 2–4. That is normal — the tool hides the factory.

🎯 Class project ideas

Project	Sources to upload
Exam revision bot	All chapter PDFs
Coaching FAQ bot	Brochure + fee sheet + timetable
Assignment helper	Syllabus + past assignment briefs
Personal RAG	Resume + skill notes

This connects directly to brochure Module 8 — AI Knowledge Assistants.

⚠️ Limits to know

Private Google account features change over time — always check current limits
Scanned blurry PDFs → bad chunks → bad answers
NotebookLM answers only from uploaded sources — it won't browse the web unless tool adds that
Still verify citations before submitting homework

✅ NotebookLM checklist

Created notebook with 3+ sources
Asked 5 questions with citations shown
Tried Audio Overview on one notebook
Found one wrong answer — traced to bad source or vague question
Wrote 1 paragraph in Obsidian: "How this is RAG"

Previous: M7-B — How RAG Works Behind the Scenes
Next: M8-A — Document-Based AI Assistants
Assignment: (Assignment) Solve Questions — long-form RAG explanations