REAL-TIME ANOMALY DETECTION MADE EASY
summary
Allegro builds a real‑time anomaly pipeline on Druid OLAP cubes with 15‑minute slices and deploys it to production, using simple percentile bounds plus an exponential moving average to forget old data and avoid overfitting to day/night cycles and deployments. Alerts use asymmetric weighting so drops trigger immediately while spikes are deprioritized. Implementation moves from R to Scala JVM services, ships to multiple projects, reduces false positives and runs in minutes rather than hours.
AI-written summary · about
chapters (6)
related talks
39mMastering real-time anomaly detection with open source tools
43mBeyond Logs: Real-Time Anomaly Detection with Machine Learning
30mBetter data quality with ML & visualization
19mThe Rise of AI in Anomaly Detection
29mAnomaly Detection
31mFirst Day Foresight: Anomaly Detection for Observability
26mAnomaly Detection System for the CERN Cloud
41mScalable Anomaly Detection (with Zero Machine Learning)