AI Help Desk - Multi-Agent System Prototype

1. Overview

This project is a prototype multi-agent AI system designed to simulate an IT Help Desk, built using the Microsoft AutoGen framework. It demonstrates agent collaboration, basic memory retention, and task execution (resolution/escalation) for handling employee IT issues as per the specified task requirements.

2. Architecture

The system employs a group chat architecture managed by a central coordinator.

• Agents:
  • User Intake Agent: Receives and processes the initial request.
  • Resolution Agent: Attempts resolution using simulated knowledge base lookups and function calls (e.g., password reset). Escalates if unable to resolve.
  • Escalation Agent: Handles escalated issues, generates a summary, simulates creating a support ticket via a function call (notify_support_team), and terminates the process.
  • Master Agent (Coordinator): (GroupChatManager) Oversees agent interactions and selects the next speaker.
  • User Proxy Agent (Internal): (UserProxyAgent) Initiates the chat internally and simulates basic user replies within the agent flow.
• Communication: Agents communicate indirectly via the GroupChatManager, which routes messages based on the conversation flow.
• Memory: The GroupChatManager maintains the conversation history for the current session, providing context and enabling resolution tracking within that session.

Diagram

Response Chat Example

How the Communication happens between Agents

• First User query pass into User Intake Agent Receives employee requests (e.g., 'I can't access my email'). Extracts key details and assigns the request to the resolution agent.

• After assigning the appropriate agent the agent try solve the query of employee for example: In our case Resolution Agent based upon the user query it will check into the knowledge base (could be any thing VectorDB or normal txt file) then give response Attempts to resolve common issues using a predefined knowledge base and function calls (e.g., resetting a password).

• If the issue is too complex, it escalates the request.Generates a structured summary for IT support and triggers a notification.

• Escalation Agent Try to solve complex query if not able to solve then Generates a ticket and raise then send to human IT Support and get involve with team and END the chat with Terminate Message.

• All agent communicate via Group chat share all information with each other and Master Agnet keep on eye on all agents.

How to Run this

Replace your OPENAI_API_KEY in .env file