Octavian Mocanu

Paul Anderson photo

Data Engineer
letgo

Had worked in academia prior to entering the private business world. Experience in test and verification, microelectronics, neuroscience (1-month stay at CalTech), low power design, GIS, programming and big data engineering. Participated and published papers in several conferences and projects.

Octavian Mocanu is speaking at the following session/s

Search and Relevance at Scale for Online Classifieds

Thursday | 1:30PM - 2:10PM |

A high performing search service implies both having an effective search infrastructure and high search relevance. Seeking for a fault tolerant, self-healing and cost-effective search infrastructure at scale, we built a platform based on Solr search engine with light compressed in-memory indexes, avoiding sharding and decreasing the overall infrastructure needs. To populate the indexes, we use flexible Spark ETL processes, keeping our product catalog and search indexes updated in a near real-time fashion and distributed across high-performant database engines. We aim at getting a high search relevance precision and recall by applying query relaxation and boost solutions on top of the optimized platform. We evaluate the responses both offline and online and finally serve them by service APIs written in highly concurrent Scala frameworks (Finatra).

Audience Takeaway
An example of a performant search platform, getting low costs and increased efficiency, speed and fault tolerance. Use performant search algorithms to increase precision and recall. Design a light, fast and flexible API taking advantage of the infrastructure and search relevance.

Intended Audience
Data engineers and data scientists.

 

Level:
All Levels

Additional speakers