Leave Management Agent

An intelligent, automated leave request processing system built with LangGraph and UiPath SDK that streamlines employee leave management through AI-powered decision making, policy compliance checks, and human-in-the-loop approvals.

Workflow

graph TD
    A[Start: Receive Leave Request] --> B[Extract Leave Information]
    B --> C[Get Employee Details]
    C --> D{All Fields Present?}
    D -->|No| E[Employee HITL Form]
    E --> D
    D -->|Yes| F[Categorize Leave Type]
    F --> G[Check Company Policy]
    G --> H[Check Leave Availability]
    H --> I{Policy Compliant?}
    I -->|Needs HR| J[HR Approval HITL]
    I -->|Auto Approved| K[Update Database]
    I -->|Auto Rejected| L[Send Rejection Email]
    J --> M{HR Decision}
    M -->|Approved| K
    M -->|Rejected| L
    K --> N[Send Approval Email]
    N --> O[End]
    L --> O

Loading

Demo Video

Features

• Intelligent Leave Request Parsing: Automatically extracts leave dates, duration, and reasons from natural language requests
• Smart Categorization: AI-powered classification of leave types (Vacation, Sick Leave, Maternity/Paternity, Bereavement, Personal Leave)
• Policy Compliance: Real-time validation against company policies using UiPath Context Grounding
• Leave Balance Management: Integration with employee database via MCP (Model Context Protocol)
• Human-in-the-Loop (HITL): Action Tasks (UiPath Action App) for missing information and HR approvals
• Automated Email Notifications: Sends approval/rejection emails with detailed information
• Database Integration: Automatic leave balance updates upon approval

Architecture

The agent is built using a state machine architecture with LangGraph, enabling complex workflows with conditional branching and human intervention points.

Key Components

1. LangGraph State Machine: Orchestrates the entire workflow
2. UiPath Services Integration:
   • Context Grounding: Policy document retrieval and compliance checking
   • MCP Server: Database operations (employee details, leave balance)
   • HITL Actions: Action Tasks (UiPath Action App) for user input and approvals
3. LLM (GPT-4): Natural language understanding and decision making
4. Email Service: Automated notifications via MCP

Getting Started

Prerequisites

• Python 3.10+
• UiPath Automation Suite account
• LLM API key
• MCP Server configured with database access

Installation

1. Clone the repository

   git clone https://github.com/rajneeshk94/LeaveManagementAgent.git
   cd LeaveManagementAgent

2. Install dependencies

   pip install -r requirements.txt

3. Configure environment variables

   Create a .env file:

   # MCP Server
   MCP_SERVER_URL=https://your-mcp-server-url
   
   # Optional: If not using UiPath's LLM
   OPENAI_API_KEY=your_openai_key

4. Set up UiPath Resources

   • Context Grounding Index: Create an index named company-policy-index with your HR policies
   • MCP Server Tools: Configure the following tools:
     • getemployeedetails: Retrieve employee information
     • getleavebalance: Fetch leave balance
     • updateleavebalance: Update leave records
     • sendemail`: Email notification service
   • HITL Apps: Deploy the following apps in your Shared folder:
     • LeaveRequestApp: Employee information app
     • HRApprovalApp: HR approval app

5. Authenticate with UiPath

   uipath auth

6. Package the project

   uipath pack

7. Publish the project

   uipath publish
   
   - When prompted, select tenant to upload

Running the Agent

import asyncio
from main import agent, GraphState

async def process_leave_request():
    # Initialize state with employee's leave request
    initial_state = GraphState(
        leave_request="I need leave from Dec 20 to Dec 25 for vacation",
        employee_email="[email protected]"
    )
    
    # Run the agent
    final_state = None
    async for event in agent.astream(initial_state):
        print(f"Event: {event}")
        final_state = event
    
    # Check final status
    if final_state:
        print(f"\nFinal Status: {final_state['final_status']}")

# Run the async function
if __name__ == "__main__":
    asyncio.run(process_leave_request())

Configuration

Leave Categories

The system supports five leave types:

• Vacation: Personal time off, travel, recreation
• Sick Leave: Illness, medical appointments, mental health
• Maternity/Paternity: Pregnancy, childbirth, adoption
• Bereavement: Death of family member, funeral attendance
• Personal Leave: Personal matters not covered elsewhere

Default Leave Balances

{
    "Vacation": 20,
    "Sick Leave": 12,
    "Maternity/Paternity": 90,
    "Bereavement": 5,
    "Personal Leave": 3
}

HR Approval Triggers

HR approval is required when:

• Leave duration exceeds 5 days
• Leave type is Maternity/Paternity
• Policy violations detected
• Insufficient leave balance
• Policy explicitly requires HR review

MCP Server Setup

Your MCP server should implement these endpoints:

GET /getemployeedetails

{
  "employee_email": "[email protected]"
}

GET /getleavebalance

{
  "employee_email": "[email protected]"
}

POST /updateleavebalance

{
  "employee_email": "[email protected]",
  "leave_type": "Vacation",
  "days": 5
}

POST /sendemail

{
  "Recipient": "[email protected]",
  "Subject": "Leave Request Status",
  "Body": "<html>...</html>"
}

Project Structure

leave-management-agent/
├── main.py                 # Main agent code
├── requirements.txt        # Python dependencies
├── .env                    # Environment variables (not committed)
├── README.md              # This file
├── Images/
│   ├── agent.png

Customization

Adding New Leave Types

1. Update the categorize_node system prompt
2. Add default balance in default_leave_balance()
3. Update company policies in Context Grounding index

Modifying Approval Rules

Edit the check_policy_node function to adjust:

• HR approval thresholds
• Policy validation logic
• Balance checking rules

Email Templates

Customize email content in:

• send_approval_email_node
• send_rejection_email_node