From keyword to vector
summary
Ten years of e-commerce work leads from managing brittle synonym lists and tokenization bugs (a 52/48 split on "wine glass" spellings) through costly experiments with learning-to-rank to adopting vector embeddings. The talk shows migrations at bull.com and Albert Heijn, a tokenization bug that returned eggs for "MMs", trials with early Weaviate prototypes, and concrete outcomes: hybrid search combining BM25 and vectors, result merging via reciprocal rank fusion (moving to relative score fusion), support for filters, images and QA, and improved Dutch multilingual embeddings.
AI-written summary · about
chapters (8)
- 0:00Introduction and personal journey
- 2:28Early days at Elastic
- 5:15Lessons from building search at Bol.com
- 10:36Challenges at Albert Heijn
- 14:51Discovery of vector search
- 16:35Transition to Weaviate
- 17:37Hybrid search explained
- 20:46Q&A session
related talks
31mBring AI based Search to Your Website
28m"Embeddings: How Computers Learned to Read"
47mBoosting AI With OpenSearch: Vector Stores, Similarity Search, and Retrieval-Augme...
30mHow to Decipher User Uncertainty with GenAI and Vector Search
48mEnter the Brave New World of GenAI with Vector Search
48mA Fun & Absurd Introduction to Vector Databases
08mRethinking your data stack: The future of AI apps and vector databases
44mRediscover your keyword search: Expand, Enrich and Rewrite