Instacart has developed a modern search infrastructure using Postgres to combine traditional full-text search with semantic retrieval, leveraging pgvector for improved performance and precision. This new system unifies retrieval mechanisms, reduces operational overhead, and enhances search relevance, accommodating dynamic inventory and complex user preferences efficiently.
Table of contents
How Instacart Built a Modern Search Infrastructure on PostgresIntroductionScale of Search Retrieval at InstacartThe Evolution of SearchOriginal ArchitectureMoving Full Text Search to PostgresSemantic SearchBringing it all together: An emergent Hybrid Search InfrastructureChoosing the right datastore for combined retrievalSort: