THE PROBLEM
Why this system existed
Keyword rules misrouted multi-topic support tickets and delayed first response across high-volume service queues.
BERT · APIs · Workflow Automation
IT Service Ticket Routing
A BERT intent classifier integrated with enterprise ticketing systems to automate routing while protecting ambiguous multi-topic requests.
MY ROLEIntent classification, confidence routing, REST integration, and operations
SCOPEA multi-class ticket-routing service for telecom and financial-services support queues with confidence-based manual escalation.
EVIDENCEProfessional experience · public-safe summary
91.2%classification accuracy
3,500+tickets per day
31%fewer misroutes
OUTCOME
What changed
Reduced misrouting from 18% to under 7% and cut mean time to first response by 22%.
REFERENCE ARCHITECTURE
Controls around the model
01Classify
02Route
03Retrieve
04Rerank
05Generate
06Validate
07Observe
DECISIONS
Trade-offs considered
Multi-class intent model
Confidence threshold routing
REST ticketing integration
Manual review for ambiguous cases
FAILURE CASE
What did not work
Low-confidence and multi-topic requests were unsafe to auto-route; the confidence tail stayed visible to human triage.
SECURITY BOUNDARY
Public-safe by design
This case study exposes patterns, not employer architecture. It uses synthetic data, no client identifiers, no internal prompts, no proprietary datasets, and no production endpoints.