LangGraph AI Agentic Systems

This repository demonstrates the use of LangGraph to build modular, graph-based AI agent systems in Python. It includes agent implementations, utilities, tutorials, and exercises for learning and extending agentic workflows.

---

Table of Contents

• Project Structure
• Installation
• Configuration
• Agents Overview
  • 1. Simple Bot
  • 2. Chatbot with Memory
  • 3. React Agent
  • 4. Drafter Agent
  • 5. RAG Agent
• Utilities
• Tutorials
• Exercises
• How to Run
• License

---

Project Structure

langgraph/
│
├── agents/
│   ├── 1_simple_bot.ipynb         # Basic conversational agent (notebook)
│   ├── 2_chatbot_with_memory.py   # Chatbot with memory (Python script)
│   ├── 3_react_agent.py           # React agent with tool use
│   ├── 4_drafter_agent.py         # Human-in-the-loop document drafter
│   ├── 5_rag_agent.py             # Retrieval-Augmented Generation agent
│   ├── utils/
│   │   ├── __init__.py
│   │   └── settings.py            # Centralized settings/configuration
│   ├── chroma.sqlite3             # Chroma vector DB for RAG agent
│   └── market-infographic-recap-2024.pdf # Example PDF for RAG agent
│
├── tutorials/                     # Step-by-step LangGraph tutorials
│   ├── 1_ hello_world.ipynb
│   ├── 2_multiple_inputs.ipynb
│   ├── 3_sequential_graph.ipynb
│   ├── 4_conditional_graph.ipynb
│   └── 5_looping_graph.ipynb
│
├── exercises/                     # Practice exercises for LangGraph
│   ├── exercise_graph1.ipynb
│   ├── exercise_graph2.ipynb
│   ├── exercise_graph3.ipynb
│   ├── exercise_graph4.ipynb
│   └── exercise_graph5.ipynb
│
├── requirements.txt               # Python dependencies
└── README.md                      # Project documentation

---

Installation

1. Clone the repository:

   git clone https://github.com/ankitprasad81/langgraph.git
   cd langgraph

2. Create and activate a virtual environment:

   python -m venv .venv
   .\.venv\Scripts\Activate

3. Install dependencies:

   pip install -r requirements.txt

---

Configuration

• Copy .env.example to .env and fill in your API keys (e.g., for Groq, HuggingFace, etc.).
• The main configuration is in agents/utils/settings.py and uses pydantic-settings.

Example .env:

GROQ_API_KEY=your_groq_api_key

---

Agents Overview

1. Simple Bot (agents/1_simple_bot.ipynb)

• A basic conversational agent using LangGraph and ChatGroq.
• Demonstrates message handling and graph compilation.

2. Chatbot with Memory (agents/2_chatbot_with_memory.py)

• Maintains full conversation history using both HumanMessage and AIMessage.
• Saves conversation logs to logging.txt.
• Run with:

  python agents\2_chatbot_with_memory.py

3. React Agent (agents/3_react_agent.py)

• Demonstrates tool use (add, subtract, multiply) within a graph.
• Uses conditional edges to decide when to call tools or finish.
• Run with:

  python agents\3_react_agent.py

4. Drafter Agent (agents/4_drafter_agent.py)

• Human-in-the-loop document drafting agent.
• Supports updating and saving documents via tool calls.
• Run with:

  python agents\4_drafter_agent.py

5. RAG Agent (agents/5_rag_agent.py)

• Retrieval-Augmented Generation agent.
• Loads a PDF, chunks it, creates a Chroma vector store, and answers questions using retrieval.
• Requires market-infographic-recap-2024.pdf in the agents/ directory.
• Run with:

  python agents\5_rag_agent.py

---

Utilities

• agents/utils/settings.py: Centralized settings using Pydantic. Reads from .env.
• agents/chroma.sqlite3: Vector database for document retrieval (used by RAG agent).

---

Tutorials

The tutorials/ folder contains step-by-step Jupyter notebooks covering:

1. Hello World: Basic graph and node setup.
2. Multiple Inputs: Handling multiple state fields.
3. Sequential Graph: Chaining nodes in sequence.
4. Conditional Graph: Routing based on state.
5. Looping Graph: Implementing loops in the graph.

Open and run these notebooks in Jupyter or VS Code to learn LangGraph concepts.

---

Exercises

The exercises/ folder provides hands-on practice with:

• State management
• Node creation
• Conditional logic
• Looping
• Multi-step workflows

Each exercise is a Jupyter notebook with a task description and starter code.

---

How to Run

• For Python scripts:

  python agents\<script_name>.py

• For Jupyter notebooks:
  Open in VS Code or JupyterLab and run cells interactively.

---

License

This project is for educational and research purposes.

---

Note:

• Make sure you have the required API keys and dependencies.
• For Windows, all commands are provided in PowerShell syntax.