Vector Search Before It Was Cool
Hard-won lessons building approximate nearest-neighbor systems for recommendations at scale.
Edition 6 · 2022
Atlanta, GAGeorgia World Congress Center
July 11–13, 2022
5,600
Attendees
104
Talks
60
Papers
Headliners
Hard-won lessons building approximate nearest-neighbor systems for recommendations at scale.
Backpressure, schema drift, and dead-letter handling for ingestion pipelines that stay calm.
How explicit producer–consumer contracts cut pipeline incidents by 60% across hundreds of teams.
Where privacy-preserving training genuinely pays off — and where a central warehouse still wins.
Bridging the online/offline gap with a single source of truth for features and labels.
Lineage, freshness, and anomaly detection — the telemetry that keeps data products reliable.
Right-sizing compute, tuning file layouts, and killing the silent spend in your data platform.
Defining freshness and completeness SLAs your consumers can plan around — and meeting them.
Proceedings
A. Okafor, T. Han
L. Zhao, V. Cruz, R. Mehta
W. Liu, J. O'Brien
P. Adeyemi, X. Dubois
J. Rivera, Z. Thompson
C. Williams, Y. Patel
G. Park, N. Gupta