Building the ultimate chatbot on your own data with Azure SQL and Semantic | Data Exposed

Learn to build with Ultimate Chatbot using RAG, NL2SQL and Semantic Kernel to query all your data, structured and unstructured. See the AI in action and understand why is going to change the way we all work and build applications, for real!

Chapters:
00:00 - Introduction
01:30 - Demo
10:39 - Getting started

✔️ Resources:

📌 Let's connect:


🔔 Subscribe to our channels for even more SQL tips:

#AzureSQL #SQL #LearnSQL

[xyola]

I did the same for where i work in elixir and postgress sql, obviously there are many things to consider like
1. make sure the sql generated is secure, correct, you want to prevent any change operations
2. In a multi-tenant single database instance, a user of one organization cannot see data of another organization
3. Performing retries
4. Avoiding db connection pool overload.
These days I recommend using SQLite as an in memory subset of data isolation if the data is not too bulky

[WalterPeirone]

This is great! I cloned and give it a try. The only thing I needed to allow null into the embedding fields because was failing while doing the run deploy. I'll play with my own data now.

[dsheardown]

Well, because I am in the UK, all I can say is CRIKEY! or BLIMEY! this is the sort of great utility AI is bringing to the business world. The company I work for has years of old support/knowledge base data - this is perfect to ask questions of it!

[sergiq]

@DavideMauri, can you suggest which model to use if I would like to play with it hosted locally ? Many companies does not want to use external API where their data are sent as a piece of prompt.

[DesignsbyBlanc]

That’s really cool! Hypothetically, if I’m working for a company and the chatbot should only surface information to a user that’s related to their respective department, what’s the ideal way to go about this? Would it involve some sort of dynamic data injection based on the user, an interface abstraction between the chatbot and the various departmental data that routes to a specific department based on the user, or would it be best to create a separate instance for each department?