Work MakingWeb Support Automation

Summary

Improved accuracy of their Zapier-based ticket auto-responder system for a customer support automation tool.

My Role

AI Engineer

Challenges

• Original dataset had mislabels and imbalance
• Client used incorrect accuracy formula
• Production system was live on Zapier

What We Did

• Re-labeled and augmented data
• Corrected accuracy calculation
• Built 3-stage LLM pipeline for classification and response

Outcomes

• Accuracy increased from 65% to 86%
• System cloned and verified without downtime
• Delivered in 3 weeks