Search at Scale: From Fuzzy Queries to OCR, Text Embedding and Reranker Models
📅 Date: Thursday, October 15 from 6:00-9:00 PM
📍 Location: Elastic London Office
Davidson Building, 1st Floor, 5 Southampton St,
London, WC2E 7HA
👉 Important: Please use the Exeter St entrance as the main entrance is closed after 6 PM.
⚠️ Entrance will be on a first come, first served basis, subject to room capacity — arrive on time to guarantee your spot.
⏰ Agenda
Arrivals, pizza and drinks: 6 PM
Welcome: 6:30 PM
Talk #1: 6:45 - 7:15 PM
Talk #2: 7:15 PM - 7:45 PM
Networking: 7:45 PM
Closing: 8:30 PM
Talk #1 by Amir Hossein Rassafi | Head of Platform at Condukt
From PostgreSQL to Elasticsearch: Migrating a Fuzzy Name Search at Scale
What happens when your search quality is good but your database can't keep up? In this talk, we'll walk through a real migration from PostgreSQL's pg_trgm to Elasticsearch — driven by growing latency and database pressure.
We'll cover how to preserve trigram-like behaviour using n-grams, prefixes, keywords, and shingles, how to validate the new system with real queries and synthetic typo cases, and the production details that made the migration safe: replication, checkpointing, lag handling, and replayable indexing.
A practical talk for anyone dealing with fuzzy search at scale — or planning a migration where quality, performance, and reliability all matter.
Talk #2 by Antonio Morais, Machine Learning Engineer | Elastic
Power of Three - turning scanned documents into ranked answers with Jina OCR, Text Embeddings and Reranker models on EIS
Take a folder of scanned documents and convert it into a searchable index with the help of three Jina models: OCR extracts the text, Text v5 embeds the text into a vector space, and Reranker v3.5 surfaces the right page at the top. We will detail the inner workings of our serving platform, and then show how you can build an end-to-end pipeline with Jina on Elastic Inference Service