Marina Svicevic

ML Software Developer, Socialgist

Marina is part of ML team at Socialgist since 2019. She has PhD in Computer Science. She is currently building the model for anomaly detection. Also, she is working on a comparison of many different methods for detecting change points in time series which we need in order to detect alarming spikes/dips in overall crawling volumes.

Talk Title:  Transforming Email Reports into Self-Serving Dashboards & Detecting Anomalies In-Stream

The first project profiled in this talk (transforming email reports into self-serving dashboards) involved generating query-based reports for the sales team replacing email reports. Messages were consumed from Apache Kafka, aggregated and inserted into InfluxDB based on Kafka metadata timestamp. The Socialgist team ran into a problem of counts mismatch for high-traffic Kafka topics and solved it by randomizing timestamp and building cache.

The second project profiled (detecting anomalies in-stream) involved detecting strange behaviors in graphs of data streams (that are internally considered as anomalies). Data is pulled from Elasticsearch, run through the anomaly detection model, and stored in InfluxDB. The results stored in InfluxDB are represented in Grafana, and alerts are fired into the Slack channel. This project helped Socialgist predict the behaviors of streams and identify alerts before any other system could.

View Schedule  >>