🤖 Building Smarter AI Agents with Google Gemini

A beginner-friendly project built during MLH Global Hack Week: Agents.

In this workshop, we use Node.js and Google Gemini to build increasingly capable AI agents — from a simple Gemini request to conversation memory, vision, and tool calling.

🚀 What You'll Learn

• Connect Node.js to the Gemini API
• Build an interactive Gemini chat agent
• Give an agent conversation memory
• Use Gemini's multimodal capabilities to understand images
• Give Gemini tools and let it decide when to use them
• Understand how function/tool calling works

🛠️ Getting Started

Clone the repository:

git clone YOUR_REPOSITORY_URL
cd ghw-agents-gemini

Install dependencies:

npm install

Create your .env file:

cp .env.example .env

Add your Gemini API key:

GEMINI_API_KEY=your_gemini_api_key_here

You can create an API key using Google AI Studio.

⚠️ Never commit your .env file or API key to GitHub.

🧪 Workshop Steps

Step 1 — Meet Gemini

Make your first request to the Gemini API.

node step-1-gemini.js

Step 2 — Chat Agent + Memory

Turn Gemini into an interactive terminal agent that remembers the current conversation.

node step-2-chat-agent.js

Step 3 — Give Your Agent Eyes 👀

Send an image to Gemini and let the agent understand and reason about what it sees.

Add your image as demo-image.jpg, then run:

node step-3-vision-agent.js

Step 4 — Give Your Agent Tools 🔧

Give Gemini access to a JavaScript weather tool and let the model decide when the tool is needed.

node step-4-tool-agent.js

Try:

What is JavaScript?

Then:

What's the weather in Lagos?

Notice that Gemini can answer the first question directly but decides to use our weather tool for the second.

🧠 How Tool Calling Works

User
  ↓
Gemini
  ↓
Decides a tool is needed
  ↓
Requests getWeather("Lagos")
  ↓
Our Node.js app runs the function
  ↓
Tool result
  ↓
Gemini
  ↓
Final answer

Gemini decides which tool to request and what arguments to provide, while our application actually executes the JavaScript function.

💡 Challenge

Take what we built and make the agent smarter!

Try:

• Adding a calculator tool
• Connecting a real weather API
• Adding another image use case
• Creating multiple tools
• Combining memory + vision + tools

🌎 MLH Global Hack Week

Session: Building Smarter AI Agents with Google Gemini

Check-in: https://events.mlh.com/events/14591-ghw-agents-week-building-smarter-ai-agents-with-google-gemini

Feedback: https://mlh.link/GHWFeedback

Happy building! 🚀