Hybrid Image Search at Scale - Lessons in Accuracy, Latency, and Cost
OpenSource Connections via YouTube
Master AI and Machine Learning: From Neural Networks to Applications
The Most Addictive Python and SQL Courses
Overview
AI, Data Science & Cloud Certificates from Google, IBM & Meta — 40% Off
One plan covers every Professional Certificate on Coursera. 40% off Coursera Plus Annual.
Unlock All Certificates
Learn how to build a scalable hybrid image search engine that serves over 10 million products by combining visual embeddings, textual signals, and knowledge-graph-enhanced metadata in this 45-minute conference talk from Haystack EU 2025. Discover the architectural decisions behind Adeo's (Leroy Merlin group) industrial-scale image search system that balances accuracy, speed, and cost-effectiveness for e-commerce applications where users show rather than describe what they want. Explore the hybrid architecture that combines image embeddings with LLM-based lexical search backed by a Knowledge Graph, and dive deep into vector quantization techniques including bf16, int8, int4, int2, and BBQ in Elasticsearch. Understand the practical trade-offs between latency, precision, and cost when implementing these quantization methods at scale, gaining insights from real-world experience in building merch-oriented search systems that capture user intent through visual queries.
Syllabus
Haystack EU 2025: Hybrid Image Search at Scale: Lessons in Accuracy, Latency, and Cost
Taught by
OpenSource Connections