← Back

Project

AI-Powered Ticket Triage System

Designed an AI-assisted triage workflow that classified incoming support requests, suggested routing and urgency, and generated structured summaries to reduce manual intake friction.

Large Language Model (LLM) APIs Prompt Design Support Workflows Structured Output Routing Logic

Focus

AI-enabled workflow design

Core outcome

Faster, more consistent triage

Approach

Human-in-the-loop AI assistance

The Problem

Incoming support requests often varied widely in clarity, urgency, and completeness, which made triage slower and less consistent than it needed to be.

Similar issues could be categorized differently depending on who reviewed them, and long ticket bodies added friction to engineering and operations workflows.

The team and partners needed a way to improve ticket intake consistency without removing human judgment from higher-risk decisions.

The Solution

I designed an AI-assisted triage layer where an LLM reviewed ticket content and returned structured outputs for classification, urgency, routing, and summarization.

Rather than replacing support review, the system served as a first-pass assistant that reduced repetitive triage work and improved consistency across intake workflows.

This created a more scalable foundation for support operations while keeping final decision-making with humans.

Workflow

From raw support request to structured triage suggestion.

1

Incoming ticket

2

LLM analysis

3

Structured output

4

Suggested routing

5

Human review

Key Features

AI-based classification

Used AI to analyze incoming support requests and assign categories based on issue type, product area, and likely ownership.

Priority suggestion

Generated an initial urgency recommendation to help teams distinguish routine requests from higher-impact issues faster.

Summary generation

Converted long-form ticket text into shorter internal summaries that were easier for triage and engineering teams to scan.

Suggested routing

Produced routing suggestions to reduce ambiguity in handoffs and improve consistency across support workflows.

Structured output design

Designed the workflow so model output could be used predictably inside operational processes instead of as unstructured text.

Human-in-the-loop review

Kept human review in the process to ensure that AI improved efficiency without replacing judgment for higher-risk decisions.

Before

  • Inconsistent categorization across tickets
  • Longer review time for verbose requests
  • More ambiguity around urgency and ownership
  • Repetitive effort during triage
  • Limited standardization across reviewers

After

  • Faster first-pass triage
  • More consistent categorization and routing
  • Easier-to-scan internal summaries
  • Reduced repetitive manual analysis
  • Human oversight preserved for final decisions

Outcomes

  • Reduced triage friction for support workflows
  • Improved intake consistency across tickets
  • Demonstrated practical AI use for operational augmentation
  • Created a foundation for future workflow automation

What I’d Improve Next

  • Confidence scoring for model outputs
  • Better routing logic tied to historical ticket data
  • Escalation triggers for higher-risk issue types
  • Feedback loop to improve prompts over time
  • Dashboarding for triage quality and model usefulness