M7-A — What is RAG?

How to read this note

RAG sounds scary. It is not. It means: before AI answers, it looks up your notes — then writes the answer using those notes.

📖 The open-book exam analogy

Closed-book exam	Open-book exam
Student uses only memory	Student can open textbook first
Might remember wrong	Can check facts in the book
ChatGPT alone	ChatGPT + your PDFs / notes

RAG = Retrieval-Augmented Generation

Retrieval = find the right pages from your material
Augmented = add those pages to the question
Generation = AI writes the answer

The real insight

RAG does not make AI smarter. It gives AI the right pages to read first — so answers come from your data, not from guessing.

🤔 Why normal ChatGPT is not enough

ChatGPT was trained on internet text — months or years ago.

Problem	What happens
Your private notes	AI never saw them
Today's fee structure	AI might invent numbers
Your company's internal policy	AI does not know it exists
Latest batch timetable	AI hallucinates dates

You need AI that answers: "According to THIS document…"

That is RAG.

🧩 RAG in one picture

     YOUR DOCUMENTS
    (PDF, notes, website)
            │
            ▼
    ┌───────────────┐
    │  Store & index │  ← happens once when you upload
    └───────┬───────┘
            │
   YOU ASK: "What is the Module 3 assignment?"
            │
            ▼
    ┌───────────────┐
    │   RETRIEVE    │  ← find relevant paragraphs
    └───────┬───────┘
            │
            ▼
    ┌───────────────┐
    │  AI GENERATES │  ← answer using retrieved text
    └───────┬───────┘
            │
            ▼
        ANSWER TO YOU

🆚 RAG vs normal chat — quick table

	Normal ChatGPT	RAG system
Knowledge source	Training memory	Your uploaded files
Best for	General questions, writing help	Company docs, course notes, manuals
Wrong answers	Hallucination risk	Lower — if retrieval finds right chunk
Setup	Zero	Upload docs + use RAG tool
Updates	Retrain model (you can't)	Upload new file (you can)

🎒 Where students use RAG

Use case	What you upload
AI Study Assistant	All chapter PDFs + your notes
Exam prep	Past question papers
Project help	Project guideline PDF
Coaching class	Brochure + syllabus
Personal	Resume + job descriptions

Brochure capstone track: "Students: AI Study Assistant" — that is a RAG project.

🛠️ Tools that do RAG (no coding needed)

Tool	What it does
NotebookLM	Upload notes → chat + podcast + slides
ChatGPT	Attach files to a Project / GPT with knowledge
Claude	Projects with document upload
Gemini	Upload to Gems / notebook features

You will build deeper systems in Module 8. For now, NotebookLM is the easiest classroom demo.

⚠️ RAG is not magic

RAG fails when:

Wrong chunk retrieved (question too vague)
Document is scanned badly (OCR garbage)
You uploaded outdated file
You never check if answer matches source

Always ask: "Which part of my document did you use?"

Good RAG tools show citations. If they don't — verify yourself.

Next: M7-B — How RAG Works Behind the Scenes — embeddings, chunks, and vector search in simple language.

Later: M8-A — Document-Based AI Assistants — building real assistants