22-25 April 2026 
Richard Conway
aizoo.info
Director at Elastacloud, Microsoft Azure MVP, Microsoft Regional Director
Video unavailable
SQLBits 2025
Fuzzy Matching with Azure OpenAI Vector Embeddings
This session will look to use new Vector Embeddings features in Azure Services, prompt engineering along with new AI patterns and practices to build AI-led fuzzy matching for Enterprise use cases. We'll look at novel non-search use cases where embeddings become a powerful and quick tool to solve a common class of problem in a new way.
With the advent of GenAI there has been a raft of development on Vector Databases both open source and Azure Vector databases. New features in SQL and Kusto make storing and comparing vectors much easier. To date most of the examples focus on using embeddings for search applications through the RAG pattern. This session will look at how to use embeddings in their own right for different use cases.
Richard Conway will walk through how to use embeddings models to build high-performance platforms to compare lists of things to get similarities matching through AI that you can't get through rules-based systems. We'll be talking about how to do this at scale with different services and patterns and practices in Azure.
In this session we'll review how to do this and write fuzzy matching platforms in Python and the kinds of workflows you'll need to construct. We'll also look at more integrated ways of doing this using Fabric and new features in SQL Server 2025 and EventHouse which operate as Vector Stores and offer some advanced features to abstract away OpenAI.
We'll be covering some interesting use cases like how we can match different systems such as CRMs in general business practice or product lists in retail which have had data keyed in incorrectly by different people over years of use and how we can align matches using embeddings.
This will be a very fast-paced informative session that will take you from beginner to hero in an under an hour.
Richard Conway will walk through how to use embeddings models to build high-performance platforms to compare lists of things to get similarities matching through AI that you can't get through rules-based systems. We'll be talking about how to do this at scale with different services and patterns and practices in Azure.
In this session we'll review how to do this and write fuzzy matching platforms in Python and the kinds of workflows you'll need to construct. We'll also look at more integrated ways of doing this using Fabric and new features in SQL Server 2025 and EventHouse which operate as Vector Stores and offer some advanced features to abstract away OpenAI.
We'll be covering some interesting use cases like how we can match different systems such as CRMs in general business practice or product lists in retail which have had data keyed in incorrectly by different people over years of use and how we can align matches using embeddings.
This will be a very fast-paced informative session that will take you from beginner to hero in an under an hour.