Demo of Azure AI & pgvector with Azure Database for PostgreSQL

Demo of Azure AI being used with Azure Database for PostgreSQL and pgvector. Based on a demo given by Claire Giordano on the keynote stage at PASS Data Community Summit, this video showcases how you can improve the relevance of search results on a recipe website by using Azure OpenAI. Specifically the demo shows how you can use the new azure_ai (Preview) extension along with the open source pgvector extension and the Azure Database for PostgreSQL managed service to deliver semantic search results-as compared to search results you could obtain in the past via pattern matching with the LIKE clause in Postgres, or via Postgres full text search.
The demo also shows how azure_ai gives you an integration between Azure Database for PostgreSQL and the Azure AI Language service, so you can do things like sentiment analysis, language detection, and PII redaction.
In combination with the Azure Database for PostgreSQL service and pgvector, the new azure_ai extension to Postgres gives you the capability to build entirely new classes of applications-entirely in Postgres.
► Video bookmarks:
⏩ 00:00 Introduction of demo
⏩ 00:25 How things worked before with LIKE
⏩ 01:09 What about full text search?
⏩ 01:45 New azure_ai extension to Postgres
⏩ 03:52 Sentiment analysis & Azure AI Language service
⏩ 04:28 Language detection
⏩ 04:47 PII redaction
⏩ 05:15 Conclusion: azure_ai & pgvector with PostgreSQL
✅ Blog post: Introducing the azure_ai extension to Azure Database for PostgreSQL: aka.ms/blog-azure-ai
📌 Docs: Azure AI extension for Azure Database for PostgreSQL - Flexible Server: aka.ms/docs-azure-ai-postgres
📌 Docs: Release notes for Azure Database for PostgreSQL - Flexible Server: aka.ms/docs-flex-postgres-rel...
✅ Follow @AzureDBPostgres on X/Twitter to stay connected: / azuredbpostgres
#PostgreSQL #AzureAI #Microsoft

[@ssrakeshsharma]

Great Feature, well presented and I really liked it.

[@yaldram-arsalan]

To the point, video with great real life demos, awesome job Ma'am, thanks a lot, will for sure try Azure db for Postgres.

[@pass4you123]

Great feature and beauty of PostgreSQL

[@hutchm92]

Excellent video!

[@smitty7326]

i DID like it! Thanks :)

[@efficiencygeek]

Could you point to a video that shows the installation and setup of the Azure AI extension, including any connection parms to the Azure OpenAI resource? Basically, an start to finish demo where all the dependencies are shown, using Postgres and an Azure account as the starting points? I understand that's quite a bit, but just relating a common starting point for devs. Still, appreciate this video as it is very eye opening to the capabilities of Postgres in Azure.

[@petropzqi]

This is great. Is a new model being created based on the database, or is the query doing all the work for each search request?

[@JoshuaJohnson-MSFT]

Thanks for the question. I'm a member of the Azure Database for PostgreSQL Product team responsible for these features.
A new ML model isn't being created in this workflow. In this example, the initial text embedding is created via a remote call to the 'text-embedding-ada-002' model, running in your Azure OpenAI Resource, from the text fields in the PostgreSQL database. The embedding is stored as a series of floating-point numbers in the vector column within the same PostgreSQL database. These vectors can then be used for similarity searches for example or as inputs in additional ML models or Azure AI services via the Azure AI extension in Azure Database for PostgreSQL Flexible Server, all from a SQL interface.
You can run vector similarity searches within Azure Database for PostgreSQL Flexible Server using the vector operators as well, Euclidean distance or cosine distance for example. More on that here: learn.microsoft.com/en-us/azure/postgresql/flexible-server/how-to-use-pgvector#vector-similarity
More information on the Azure AI extension can be found here : learn.microsoft.com/en-us/azure/postgresql/flexible-server/generative-ai-azure-overview

[@petropzqi]

​​@@JoshuaJohnson-MSFTthx for the answer and the resources

[@sultonbekrakhimov6623]

Awesome, I recently learned about Semantic Kernel project by Microsoft, and it makes this adding features to existing data access flow so much easier, as everything is done within the code, so we can add Open AI connector and bring Language Services to do sentiment analysis, the big difference and benefit from that I see, maintaining extensions and changing underlying models for extensions for PgSQL server would be much more painful that changing it in infrastructure, inside code
So I prefer code based approach on building systems with extended AI features

[@TheUsamawahabkhan]

Great

[@Tony-dp1rl]

That is one expensive application to run at any sort of scale by the time you pay for all those Azure resources.