Detecting anomalies in Flipkart's fulfillment network
summary
Flipkart moves from dashboards to an automated anomaly pipeline that samples metrics, stores them in an incremental data platform, and scores anomalies with Isolation Forest adapted to time series; models run per metric, use ElasticSearch as a terminal store, and operate with a 20‑minute SLA for visibility. The system handles lakhs of shipments and high variance, yields measurable cost savings, starts unsupervised with human acknowledgement as feedback, and plans to collect labels to train supervised models and automate actions.
AI-written summary · about
chapters (10)
- 0:00Logistics and aviation
- 2:30Flipkart fulfillment
- 5:11Control tower challenges
- 7:40Solution evolution
- 9:29Defining the problem
- 12:01ML algorithm implementation
- 14:40Data processing platform
- 17:12Key learnings and insights
- 20:56Future automation goals
- 23:39Q&A session
related talks
30mBetter data quality with ML & visualization
20mSmart Monitoring System for Anomaly Detection on Business Trends
29mIntelligent Log Analysis and Real-time Anomaly Detection @ Salesforce
43mBeyond Logs: Real-Time Anomaly Detection with Machine Learning
21mAnomaly Detection for Platform Logs using Embedded Context
38mWatch Everything, Watch Anything: Anomaly Detection
26mAnomaly Detection System for the CERN Cloud
31mFirst Day Foresight: Anomaly Detection for Observability