This talk introduces Yelp’s new Elasticsearch-based search backend that serves billions of queries per year for tens of millions of unique users per month. In particular we discuss the challenges of transferring our existing complex custom Lucene-based ranking and text analysis functionality to Elasticsearch. We also highlight several new technologies such as our contributions to Learning to Rank that allow us to host machine learning models directly in Elasticsearch, as well as our Maptype plugin that enables our scoring logic to efficiently lookup sparse map-type data during query time.

Speaker info

Umesh Dangat

Umesh Dangat is a Group Tech Lead at Yelp. He has helped modernize Yelp’s search infrastructure and moved it from a custom distributed Lucene-based ranking application to a generalized ranking application built on top of Elasticsearch. He has also architected and implemented a custom geocoder for Yelp to mitigate reliance on third-party geocoders. This system serves the majority of the geocoding traffic at Yelp today. He currently works on building a ranking platform at Yelp which enables multiple teams at Yelp to quickly deploy their machine learned models on elasticsearch for customized scoring. He is an open source contributor to Elasticsearch and Learning to Rank plugin for Elasticsearch.

Guenther Starnberger

Guenther Starnberger is a Software Engineer at Yelp. As Search Infrastructure Tech Lead, he architected and implemented a new microservice-based search stack using Elasticsearch as the primary backend. Prior to that he worked as a Research Assistant at the Distributed Systems Group of the Vienna University of Technology where he also received his doctorate. He currently works on integrating Apache Beam into Yelp’s streaming infrastructure.

Karthik Alle

Karthik Alle is a Software Engineer on the Ranking Platform Team at Yelp. He has pioneered Search Personalization efforts and has worked on moving Yelp's legacy search system to an Elasticsearch-powered system. He has also led multiple projects in designing and implementing Indexing, Relevance and Ranking frameworks on top of Elasticsearch. Currently, he is leading the modernization efforts on the Platform to a flink-based indexing framework.