Search on Steroids
summary
Combines lexical search (Elasticsearch) with vector search using a Hugging Face model to vectorize queries and products, storing vectors in Elasticsearch as a vector database. Results from both searches are fused with Reciprocal Rank Fusion to return the top-5 nearest neighbors, improving synonym and contextual relevance while preserving exact matches; deployed on a 70+ brand catalog. Notes trade-offs: duplicated pipelines, increased latency and operational cost, and the need for careful model selection and high-quality data.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:27Lexical search limits
- 3:54Vector search explained
- 6:03Vectorization integration
- 7:10Vectorization model choice
- 8:11Vector search results
- 9:13Hybrid search and RRF
- 10:49Hybrid result fusion
- 12:35Production benefits
- 12:55Solution limits and costs
- 13:39Conclusion
related talks
19mVectorize Your Open Source Search Engine
01h 01mVector Similarity Search in Spring
45mRevolutionizing Search with Power of Neural Sparse and Dense Vectors
01h 48mInformation Retrieval from the Ground Up
01h 01mVector Similarity Search in Spring with Redis Stack
55mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
47mWhy, What, When and How of Running Diff Types of Sear...
27mMachine Learning for Advanced AI Search