Learning Langchain with Next.js & MongoDB: HR Chatbot

This project documents my learning journey exploring Langchain, MongoDB Atlas Vector Search, and Next.js by building a simple HR assistant chatbot. The goal was to understand Retrieval-Augmented Generation (RAG), agent creation, state management with LangGraph, and integrating various AI models and databases within a modern web framework.

✨ Features

• Chat Interface: A simple UI built with Next.js and Tailwind CSS to interact with the HR agent.
• HR Agent: A Langchain agent capable of answering questions based on employee data.
• Employee Data Lookup: Uses MongoDB Atlas Vector Search to find relevant employee information based on natural language queries.
• Database Seeding: A script (seed-database.ts) to generate synthetic employee data and populate the vector database using an LLM.
• Model Flexibility: Designed to potentially swap between different LLMs (OpenAI, HuggingFace, Mistral, Ollama) and embedding models.
• State Management: Uses LangGraph with MongoDB persistence to manage conversation history.

📚 Tech Stack

• Framework: Next.js (App Router)
• Language: TypeScript
• AI/LLM Orchestration: Langchain & LangGraph
• Database: MongoDB Atlas (with Vector Search)
• Styling: Tailwind CSS
• LLM (Primary): OpenAI (Configurable for others like Mistral, HuggingFace, Ollama)
• Embeddings (Primary): OpenAI (Configurable for others like HuggingFace)
• Package Manager: pnpm

🚀 Getting Started

Follow these steps to set up and run the project locally.

Prerequisites

• Node.js (LTS version recommended)
• pnpm
• A MongoDB Atlas account with a running cluster.
• API Keys:
  • OpenAI API Key
  • (Optional: Mistral API Key, HuggingFace API Key if configuring different models)

Installation

1. Clone the repository:

   git clone <your-repo-url>
   cd <repo-name>

2. Install dependencies:

   pnpm install

Environment Variables

1. Copy the sample environment file:

   cp .env.sample .env

2. Populate the .env file with your credentials:

   # MongoDB Atlas connection string (make sure IP access is configured)
   MONGODB_URI=mongodb+srv://...
   
   # OpenAI API Key
   OPENAI_API_KEY=sk-...
   
   # Optional: Set to true to clear the database before seeding
   RESET_DB=false
   
   # Optional: Add other keys if using different models
   # MISTRAL_API_KEY=...
   # HUGGINGFACE_API_KEY=hf_...

Database Setup & Seeding

1. Configure MongoDB Atlas:

   • Ensure you have created a database and collection (names specified in src/app/utils/config.ts).
   • Create a Vector Search Index on your collection (details in src/app/utils/config.ts for index name and fields).

2. Seed the database: This script uses the configured LLM (currently OpenAI) to generate synthetic employee data and creates embeddings.

   pnpm seed-db

   • Note: If you want to clear existing data before seeding, set RESET_DB=true in your .env file.

Running the Application

1. Start the development server:

   pnpm dev

2. Open http://localhost:3000 in your browser.

🛠️ Project Structure (Key Directories)

💡 Key Learnings & Concepts Explored

• Retrieval-Augmented Generation (RAG): Combining LLM capabilities with external knowledge retrieval from a vector database.
• Langchain Agents: Building autonomous agents that can use tools (like database lookups) to answer questions.
• LangGraph: Managing multi-step agent interactions and conversation state.
• MongoDB Atlas Vector Search: Setting up and querying a vector database for semantic similarity search.
• Next.js App Router: Building API routes and server/client components.
• Environment Management: Handling API keys and configuration securely.
• Database Seeding: Programmatically generating and populating database records.
• Model Integration: Swapping between different LLM providers and embedding models within Langchain.

License

This project is licensed under the MIT License - see the LICENSE file for details.