Using Dense Vector search at the EU Publications Office
summary
Proof of concept adds Elasticsearch dense vector support plus Google BERT (Hugging Face) embeddings, distilBERT QA and Pureinsights Discovery Platform to a corpus of 200k+ publications and a law dataset of ~1M records, chunking long docs to ~512 tokens, storing vectors and running cosine/DOT similarity to surface extractive answers and highlight exact dates or figures. A candidate-selection and re-ranking stage chooses the best snippet, demo returns Google‑like featured snippets and related documents, but multilingual coverage is limited to English, French, Spanish.
AI-written summary · about
chapters (10)
- 0:00Introduction to proof of concept
- 1:03What is semantic search
- 4:52Publications Office mission
- 8:46Trend toward natural language queries
- 10:13Emulating Google with Elasticsearch
- 11:18Google search examples
- 15:30Proof of concept demo
- 18:49How the technology works
- 25:39Platform architecture overview
- 29:15Q&A session
related talks
30mHow to Decipher User Uncertainty with GenAI and Vector Search
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!
15mUnconference: Explainable Vector Search
55mSupercharged Search with Semantic Search and Vector Embeddings
22mVector Search Made Simple: Getting Started With OpenSearch for AI Applications
48mBuilding an AI/ML powered text search system
18m: Workforce Efficiency with AI-Driven Vector Search
02h 51mBuilding Your Intelligent Assistant with Hugging Face and Elasticsearch