A naukri.com initiative
Analyticsindiamag
6d
405
Image Credit: Analyticsindiamag
Why Instacart Moved to Postgres & pgvector to Boost Semantic Search
- Instacart, serving 14 million daily users, faced challenges in delivering fast and accurate semantic search beyond keyword matching.
- Previously using Elasticsearch and FAISS, Instacart transitioned to Postgres and pgvector for enhanced search performance.
- The migration to Postgres, with a highly normalized data model, reduced write workload significantly for Instacart.
- Postgres allowed for storing ML features separately, providing flexibility for more sophisticated retrieval models.
- By moving compute closer to storage using Postgres on NVMe, search performance for Instacart doubled.
- Instacart's migration to pgvector from FAISS unified retrieval mechanisms, reducing operational complexity and improving search quality.
- Several companies, including Shopify, have adopted modern search infrastructures to enhance search capabilities and consumer intent understanding.
- Shopify improved search intent with real-time machine learning capabilities, focusing on understanding consumer intent beyond keyword matching.
- Shopify's AI-powered semantic capabilities processed 2,500 embeddings per second on Google Cloud Dataflow, optimizing for up-to-date embeddings for improved sales and user experience.
- Shopify's solution involved multiple model copies to keep GPUs busy while balancing trade-offs between efficiency and latency for real-time image processing.
Read Full Article
24 Likes
For uninterrupted reading, download the app