Advancing Search at Uber with OpenSearch
summary
Uber runs real-time search at massive scale: over 15 billion queries a day, indexing one million+ restaurants and serving one million concurrent trips. The platform extends OpenSearch with semantic vector search and LLM-generated embeddings, a query-time embedding retrieval plus second-pass ranker, and context signals (time, location) to boost relevance and revenue. Ingest pipeline reads from Kafka via pull-based ingestion (OpenSearch 3.0) and uses virtual shards, cloud‑native cluster state, and gRPC/protobuf transport (≈50% faster on large payloads). Contributions go upstream to OpenSearch.
AI-written summary · about
chapters (11)
- 0:00Uber’s search platform intro
- 1:07How search powers Uber
- 3:38Real-time indexing demands
- 5:28Transition to semantic search
- 8:18Modern search architecture
- 10:01Why Uber embraced OpenSearch
- 11:28Uber’s OpenSearch contributions
- 13:55Cloud-native OpenSearch
- 14:18Kafka-based ingestion
- 15:44gRPC and Protobuf for OpenSearch
- 16:49Open-source ecosystem innovations
related talks
15mAdvancing Search and Observability at Uber with OpenSearch
24mVector Search at Uber
26mAdvancing Multi-Modal Search Capabilities in Search Pipeline
41mIntegrating Realtime User Behavioral Embeddings Into OpenSearch
41mText to Context: How We Introduced Hybrid Search
26mOpenSearch and the Path to a Trillion-s...
48mUnlocking Next-gen Semantic Search With OpenSearch 3.x
40mBoost your workloads with OpenSearch innovations