JS AI Agent using LangGraph and MongoDB 🚀🤖

This project showcases how to integrate LangGraph with MongoDB for building and managing AI agents and conversational applications. I created this project to explore the integration of language models, graph-based conversation management, and MongoDB for data persistence, enabling the creation of intelligent, autonomous agents using TypeScript and Express.js.💡💻

Features ✨

• LangGraph Integration: Manages agentic conversational flows in TypeScript. 🔄
• MongoDB Atlas: Stores and retrieves conversation data.☁️🗄️
• RESTful API: Built with Express.js for handling chat interactions.🌐
• AI Integration: Utilises Google's Gemini for generating responses and embeddings.🤖🧠
• Fullstack Principles Lookup: Implements MongoDB Atlas vector search for retrieving and discussing fullstack engineering principles. 🔍🧱

Prerequisites

• Node.js and npm
• A MongoDB Atlas account
• Google AI (Gemini) API key

Installation 🛠️

1. Clone this repository:

   git clone https://github.com/andrenormanlang/javascript-ai-agent
   cd javascript-ai-agent

2. Install dependencies:

    npm install

3. Set environment variables:

   Create a .env file in the root directory and add the following:

   GOOGLE_API_KEY=your_gemini_api_key_here
   MONGODB_ATLAS_URI=your_mongodb_atlas_uri_here

   Optional:

   GEMINI_MODEL=gemini-3-flash-preview
   GEMINI_EMBEDDING_MODEL=embedding-001

4. Seeding the Database:

cd backend npx ts-node database-seed.ts ```

5. Atlas Vector Search Indexing:

Go to your MongoDB Atlas dashboard and create a new vector search index for the fullstack_db.principles collection as a JSON editor.

Index Name: vector_index

Index Definition:

```json
{
  "fields": [
    {
      "numDimensions": 768,
      "path": "embedding",
      "similarity": "cosine",
      "type": "vector"
    }
  ]
}
```

This index will be used for retrieving fullstack principles based on their embeddings.

Note: if you change `GEMINI_EMBEDDING_MODEL`, the embedding dimension may change.
The seed script logs the actual embedding length it gets from Gemini—match `numDimensions` to that value.

Usage ▶️

1. Start the server:

   npx ts-node index.ts

2. API Endpoints:

Start a new conversation 💬

curl -X POST -H "Content-Type: application/json" -d '{"message": "Your message here"}' http://localhost:3000/chat

• Curl example:

curl -X POST   -H "Content-Type: application/json"   -d '{

    "message": "How can I make a responsive layout that is also accessible for people with disabilities?"
  }'   http://localhost:3000/chat

Continue an existing conversation 🔄

curl -X POST -H "Content-Type: application/json" -d '{"message": "Your follow-up message"}' http://localhost:3000/chat/{threadId}

• Curl example:

curl -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "message": "Can you tell me what are some common pitfalls when implementing these practices?",
    "thread_id": "1739739439130"
  }' \
  http://localhost:3000/chat/1739739439130

Project Structure 📁

• backend/index.ts: Entry point for the Express.js server and API routes.
• backend/agent.ts: Defines the LangGraph agent, its tools and the conversation flow.
• backend/database-seed.ts: Script for seeding MongoDB Atlas with synthetic fullstack principles + embeddings.

How it Works ⚙️

• Data Seeding: The backend/database-seed.ts script generates synthetic fullstack principles and populates MongoDB with embeddings for Atlas Vector Search. 🌱
• LangGraph Agent: Defined in agent.ts, it manages the conversation graph structure and integrates the necessary tools. 🔧
• Database Integration: MongoDB operations are directly integrated into the agent for storing and retrieving conversation data. 💾
• API Endpoints: The Express server in index.ts provides endpoints for starting and continuing conversations. 📡
• State Persistence: Conversation data is persisted in MongoDB Atlas, ensuring continuity across sessions. 🔒