Customer Support Ticket Triage Agent

Problem

Slow customer response due to manual categorisation of support tickets.

Solution

• Built a classification agent that classifies incoming messages and predicts intent using a fine-tuned LLM.
• Assigns urgency using a formula: urgency score = sentiment × topic weight.
• Automated routing to the right support specialist.
• Drafts first responses for common issues with human approval in the loop.
• Designed SOP for the support team to oversee and escalate.

Outcome

• Response time reduced from forty hours to four minutes.
• Support backlog reduced by eighty five percent.
• Customer satisfaction improved materially.