TingIS: Real-time Risk Event Discovery from Noisy Customer Incidents at Enterprise Scale
Researchers have developed TingIS, a novel system designed to identify critical technical issues in real-time from noisy customer feedback. The system employs a multi-stage engine that combines Large Language Models with efficient indexing to merge and extract actionable incidents from user descriptions. TingIS also incorporates a cascaded routing mechanism for business attribution and a noise reduction pipeline, achieving a 95% discovery rate for high-priority incidents with a P90 alert latency of 3.5 minutes in a production environment. AI
IMPACT This system could significantly reduce downtime for cloud services by enabling faster discovery and mitigation of critical technical issues.