AI-powered Help Desk Agent

Agent Description:

The AI-powered Help Desk Agent automates first-line support by ingesting IT tickets via an API. It classifies each ticket by category (Hardware, Software, etc.) and extracts relevant keywords. The agent then searches an SQL knowledge base for a matching solution based on these details. If a relevant article is found, it provides the solution steps ("Resolved"); otherwise, it marks the ticket for escalation.

Purpose and Components
  • Purpose: To provide an automated first-line support system that can instantly resolve common issues by matching incoming tickets to a pre-defined SQL knowledge base, reducing manual effort and improving response times.
  • Components:

    • API Connector (GET): A tool to ingest new support tickets from an external source.

    • Ticket Classifier: An agent that analyzes and categorizes incoming tickets.

    • Knowledge Base Searcher: An agent that queries an SQL database to find solutions.

    • SQL Database Toolkit: A tool to connect and interact with the (SQLite) knowledge base.

Supported Capabilities

  • Receiving new support tickets via a GET API request.

  • Extracting and parsing ticket details (e.g., ticket_id, subject, keywords).

  • Standardizing data, such as lowercasing and de-duplicating keywords.

  • Classifying tickets into predefined categories (Hardware, Software, Network, Account).

  • Querying an SQL knowledge base (SupportTicketsSolutions) for solutions.

  • Matching ticket categories and keywords to knowledge base article titles.

  • Automatically marking tickets as "Resolved" and providing solution steps if a match is found.

  • Automatically marking tickets as "Escalated" if no matching solution is found.

  • Outputting a final structured JSON response with the ticket's status and solution (if any).

LLM Used

  • GPT_4O_MINI

    Note: To learn more about the LLM and to modify its behavior, refer to the Configuring LLM settings section.

Sub-Agents

1. Ticket Classifier Subagent

  • Role: Ticket Classifier

  • Scope: Analyze incoming support tickets and classify them into predefined categories (Software, Hardware, Network, Account) with priority and keywords.

  • Description: This agent serves as the entry point. It uses the GET tool to read an incoming support ticket (in JSON format). It then extracts key fields, standardizes the keywords (e.g., lowercase, remove duplicates), and assigns a primary category (Hardware, Software, etc.). Finally, it passes this structured and classified ticket data to the Knowledge Search Subagent.

  • LLM Used: Default (Inherits from parent agent).

  • Tools Used: Request - Get

2. Knowledge Search Subagent

  • Role: Database Searcher

  • Scope: Query the SupportTicketsSolutions SQL table to find matching solutions based on category and keywords.

  • Description: This agent receives the classified ticket data. It then uses the SQL toolkit to query the SupportTicketsSolutions database. It compares the ticket's category and keywords against the database's Category and Title fields. If a relevant article is found, it marks the ticket as "Resolved" and provides the solution steps. If no match is found, it marks the ticket as "Escalated" to be forwarded to a support team.

  • LLM Used: Default (Inherits from parent agent).

  • Tools Used: SQL - Toolkit
Tools Used:

  • Request - Get Tool – Fetches the incoming support ticket data from a JSON endpoint.

  • SQL - Toolkit (SQLite) – Executes queries against a SQLite database containing the SupportTicketsSolutions knowledge base.

Note: For details on modifying the Tools, refer Tools Library section.
Agent Workflow Behavior Summary

  1. The Ticket Classifier Subagent (start node) is triggered and uses the Request - Get tool to fetch a new support ticket.

  2. It parses the ticket, assigns a category (e.g., "Hardware"), and cleans up the keywords.

  3. This structured data is passed to the Knowledge Search Subagent.

  4. The Knowledge Search Subagent (end node) uses the SQL - Toolkit to search the database for an article matching the ticket's category and keywords.

  5. It generates a final output: if a match is found, it returns the solution steps and a "Resolved" status. If no match is found, it returns an "Escalated" status.

Sample Questions:

  • I am unable to log in to the portal

  • How do I install the new software update?