AIOps: New Approaches to Reduce Alert Noise
summary
Claim: an embedding‑based pipeline cuts alert noise dramatically—93% reduction in alert volume and response time improved by over four hours. Implementation uses embedding similarity search with text embeddings (Hugging Face), vectors stored in a vector database, and a SaaS alerting platform that demos collapsing 31 Prometheus floods into ~5 alerts. Also prototypes agentic root‑cause analysis that reads logs, deployments and GitHub diffs to identify a Stripe error and a bad commit. cloud only product; on‑prem rebuild possible.
AI-written summary · about
chapters (5)
- 0:00Introduction
- 3:00Alert overload harms ops
- 12:40Demo: alert deduplication
- 23:03Future of incident response
- 35:46Q&A
related talks
29mAI-Powered Anomaly Detection in Cloud Data Pipelines
25mMonitoring AI Pipelines Output As Product
42mMachine Learning In Observability Systems: From Insights To Action
40mNext-generation Alerting and Fault Detection
11mTransforming Problem-Solving with Data-Driven Insights
12mUnleashing the Power of AI for Performance Monitoring& Anomaly Detection
37mAIOps: Anomaly Detection with Prometheus and Istio
19mAIOps