RAG Explained

Ever wondered how ChatGPT could answer questions about your documents? That's RAG. This tutorial breaks it down into simple, runnable code that actually works.

The Files

• what_is_RAG.md - Start here if RAG is new to you
• How_does_RAG_work.ipynb - Interactive notebook (run each cell)
• Complete_RAG_implementation.py - The full system (ready to run)
• knowledge.txt - Your test data
• RAG_knowledge.md - Advanced stuff and FAQs

---

Contents

• Quick Start
• See It in Action
• How to Use This Repo
• What Makes This Different
• The 5-Step Breakdown
• What is RAG?
• Resources

---

Get Running

Step 1: Grab the code

git clone https://github.com/Makilesh/RAG_Explained.git
cd RAG_Explained

Step 2: Install stuff

python -m venv .venv
.venv\Scripts\activate          # Windows
source .venv/bin/activate       # Mac/Linux
pip install -r requirements.txt

Step 3: (Optional) Get a free Gemini API key

• Go to Google AI Studio
• Copy .env.example to .env and paste your key
• Skip this if you want to run locally

Step 4: Run it

python Complete_RAG_implementation.py

Ask it: "What do elephants use their trunks for?"

See It in Action

Here's what happens when you run it:

$ python Complete_RAG_implementation.py

✅ Loaded documents: 9 chunks

Enter your question: What do elephants use their trunks for?

Retrieved: Elephants use their trunks for smelling, breathing, trumpeting, drinking, and grabbing things.

Augmented Prompt:
Based on this info: Elephants use their trunks for smelling, breathing, trumpeting, drinking, and grabbing things.
Answer the question: What do elephants use their trunks for?

Use Gemini API for generation? (y/n, default n): y

Final Answer (Gemini):
Elephants use their trunks for smelling, breathing, trumpeting, drinking, and grabbing things.

---

How to Use This Repo

Never touched RAG before?
Read what_is_RAG.md → Open How_does_RAG_work.ipynb → Run each cell and watch what happens

Just want to see it work?
Run Complete_RAG_implementation.py right now. It works out of the box.

Building something real?
Check RAG_knowledge.md for production tips, common pitfalls, and how to scale.

What Makes This Different

No fluff. No 500-line enterprise boilerplate. Just:

• Real code that runs in 5 minutes
• Actual explanations of why each line exists
• Works offline (uses local models if you skip the API)
• Your choice: Gemini API for quality, or free local model

Built with: LangChain, FAISS, Sentence Transformers, Google Gemini

The 5-Step Breakdown

This is what RAG actually does:

1. Ingest → Load your documents (PDFs, text files, whatever)
2. Embed → Turn text into numbers (so computers can compare them)
3. Retrieve → Find the most relevant chunks for your question
4. Augment → Combine your question with the retrieved info
5. Generate → Let an LLM write a proper answer

Each step has its own section in the notebook with runnable code.

Try These Questions

Once you're running, ask:

• "What do elephants use their trunks for?"
• "Tell me about honey"
• Or add your own docs to knowledge.txt and ask about those

What is RAG, Really?

Full explanation in what_is_RAG.md →

You know how ChatGPT sometimes makes stuff up? RAG fixes that by letting it "look up" answers in your documents before responding.

Without RAG: LLM guesses based on training data
With RAG: LLM reads your actual documents, then answers

It's like giving someone an open-book exam instead of making them memorize everything.

Issues? Questions?

Open an issue. I actually read them.

Or fork it and make it better. That works too.

If this helped you build something, drop a star ⭐

Resources that helped build this:

• LangChain Docs
• FAISS on GitHub
• Gemini API
• Sentence Transformers