LangGraph Chatbot Workflows

A simple LangGraph-based project that demonstrates how to build LLM workflows using graph nodes, shared state, and conditional routing.

Features

• Basic LangGraph chatbot flow
• State-based message handling
• Multiple graph nodes
• Conditional routing between nodes
• OpenAI gpt-4.1-mini integration
• .env support for API keys

Tech Stack

• Python
• LangGraph
• LangChain
• OpenAI API
• python-dotenv

Project Structure

Langraph/
├── chat.py      # Basic LangGraph flow with chatbot and sample node
└── chat2.py     # Conditional graph flow with response evaluation

How It Works

chat.py

This file creates a simple graph:

START → chatbot → samplenode → END

The graph takes an initial message, passes it through a chatbot node, appends another message through a sample node, and returns the updated state.

chat2.py

This file demonstrates a slightly more advanced workflow:

START → chatbot → evaluate_response → endnode → END

It sends the user query to an OpenAI model, stores the LLM response in state, and uses conditional routing to decide the next node.

Setup

Clone the repository:

git clone https://github.com/pranjalisr/Langraph.git
cd Langraph/Lang/Langraph

Install dependencies:

pip install langgraph langchain openai python-dotenv typing-extensions

Create a .env file:

OPENAI_API_KEY=your_openai_api_key_here

Run the Project

Run the basic graph:

python chat.py

Run the conditional graph:

python chat2.py

Example Query

updated_state = graph.invoke({
    "user_query": "Hey, What is 2+2?"
})

What I Learned

• How LangGraph represents AI workflows as graphs
• How to define shared state using TypedDict
• How to add nodes and edges in a LangGraph workflow
• How to use conditional edges for routing
• How to connect OpenAI models with graph-based execution
• Future Improvements
• Add proper response quality evaluation
• Add Gemini or another fallback model
• Add memory support
• Add streaming responses
• Build a simple UI using Streamlit or FastAPI
• Add error handling and logging