Build Your First RAG System with Gemini API

🤔 What Even Is RAG?

Imagine you have a super-smart AI assistant but instead of guessing answers from memory, it reads your notes first, then answers based on your facts. That’s RAG: Retrieval-Augmented Generation.

• Retrieval: Find relevant info from your documents
• Augmentation: Give that info to the AI
• Generation: Let the AI write a smart, grounded answer

Today, you’ll build a tiny RAG system using Google’s free Gemini API no servers, no credit card, and under 100 lines of code.

---

🛠️ What You’ll Need

1. A Google account (to use Gemini)
2. Python installed (v3.8+)
3. 5 minutes of your time 😊

💡 Don’t worry if you’ve never used AI APIs before we’ll walk through everything!

---

🔑 Step 1: Get Your Gemini API Key

1. Go to Google AI Studio
2. Click “Get API key” (top-right)
3. Copy the key (it looks like AIzaSy...)
4. Save it somewhere safe (you’ll use it in a minute)

✅ Free tier: Gemini gives you 60+ requests per minute for free plenty for learning!

---

📁 Step 2: Create Your “Knowledge” File

RAG needs something to read! Let’s make a simple text file.

Create a file called facts.txt and paste this:

Hacktoberfest is a month-long celebration of open source software.
It started in 2014 by DigitalOcean.
To earn a T-shirt, you need 4 valid pull requests in October.
It’s beginner-friendly and encourages new contributors.

This is your private knowledge base the AI will only use this to answer questions.

---

🐍 Step 3: Install Python Tools

Open your terminal and run:

pip install google-generativeai chromadb sentence-transformers

What these do:

• google-generativeai: Talks to Gemini
• chromadb: Stores document “memories” (a vector database)
• sentence-transformers: Turns text into numbers AI can compare

---

🧩 Step 4: The Magic Script (rag.py)

Create a new file called rag.py and paste this:

import os
from google.generativeai import GenerativeModel
import google.generativeai as genai
from chromadb import PersistentClient
from sentence_transformers import SentenceTransformer

# === 1. SET UP GEMINI ===
genai.configure(api_key="YOUR_API_KEY_HERE")  # 👈 PASTE YOUR KEY!
model = GenerativeModel("gemini-1.5-flash")

# === 2. LOAD YOUR FACTS ===
with open("facts.txt", "r") as f:
    facts = f.read().split("\n\n")  # Split into chunks

# === 3. TURN TEXT INTO "MEMORY" (EMBEDDINGS) ===
embedder = SentenceTransformer("all-MiniLM-L6-v2")
embeddings = embedder.encode(facts)

# === 4. SAVE TO VECTOR DATABASE ===
client = PersistentClient(path="./my_rag_db")
collection = client.get_or_create_collection("facts")
collection.add(
    embeddings=embeddings.tolist(),
    documents=facts,
    ids=[f"fact_{i}" for i in range(len(facts))]
)

# === 5. ASK A QUESTION! ===
question = "When did Hacktoberfest start?"
print(f"❓ Question: {question}\n")

# Find the most relevant fact
results = collection.query(
    query_embeddings=embedder.encode([question]).tolist(),
    n_results=1
)
context = results["documents"][0][0]

# Ask Gemini using your fact
prompt = f"Answer using only this context:\n{context}\n\nQuestion: {question}"
response = model.generate_content(prompt)

print(f"🤖 Answer: {response.text}")

🔑 Don’t forget: Replace "YOUR_API_KEY_HERE" with your actual Gemini key!

---

▶️ Step 5: Run It!

In your terminal, run:

python rag.py

You should see:

❓ Question: When did Hacktoberfest start?

🤖 Answer: Hacktoberfest started in 2014.

✅ It worked! The AI didn’t guess it used your file to answer.

---

🌟 Why This Is Cool

• No hallucinations: Answers come from your data
• Private: Your facts.txt never leaves your computer
• Scalable: Add more files, PDFs, or web pages later
• Free: Uses free-tier APIs and open-source tools

---