LLM-Powered AI Dashboards from Databricks One

0:04Bro. Hey folks. Today I have the pleasure to host Kristoff Shu. He's a dear colleague. But before introducing him, he came up with an amazing idea. I think for those who are using now databicks one, the consumer mode released by databicks where we can see the dashboards, apps and genie. Well, some folks asked whether we can build a

0:29dashboard directly from datab bricks one and be able to consume it. I have good news for you. Kristoff was able to make your dream come true. >> Hi Kristoff. >> Hi Yousef. [snorts] Thank you for having me today. >> Can you introduce yourself? >> Yes. Um hi everyone. I'm Kristoff. I'm a colleague of you. I'm a senior engineer at Databricks for almost two years now.

0:51Yeah. Happy to show today what we have at datab bricks >> and can you tell us or maybe show us how were you able to make it? >> Yes. And maybe okay then maybe let me give you a demo and then show directly what I we have um so now I am uh a data analyst or business analyst and I

1:14have access to data one. So uh I have my engineers who gave me some dashboard but I want to have my own dashboard. what should I do and this is what we develop a datab bricks so this is an app on which uh you have now a new way to create a consume dashboard using an LLM on our case it's TGPT5 available with a

1:37foundational mod API how does it work let's say I'm working uh in the customer support department so I'm going to log my data uh from the customer support department um that we have in UT catalog and hey let's say here I am an analyst for the customer support department and

2:00I am looking to understand the performance of my department. So now we're going to uh submit this prompt to the LLM is going to uh take a look look through the different um widgets available for the dashboard. select the most relevant one and here

2:22you can see we can follow the reasoning behind to understand better why he picked uh these widgets and why it made sense uh for him and then once we have it we have our dashboard that is deployed in datab bricks and then embedded as an iframe into apps and then now it's loading loading and we should have normally not only thing

2:47and yes not everything is here but we have at least three main widgets which is the table which show us the level the level of detail that we need to understand the different records registered for each uh customer support agent. We have here the number of tickets reserved by the first line support and also by second line support

3:09when the ticket is escalated and here we can understand better uh the number of tickets we got through the different days per priority is it low priority, high priority or medium priority. Um just to understand like when the AI build or the LLM build the dashboard is it automatically published or is it a draft mode? It's automatic publish. Um that's a very good question. Let me show you now. If we go to our latest one

3:34and we go back to our homepage and now we go to dashboard and we have here our dashboard published just to show you we have the same name per view and then this is generally this is randomly generated just to ensure that we don't have the same name every time generated which will create a conflict and lead to an issue. So here we have our dashboard and then if I click on here then we will

3:56see that it is published and then I can always adjust it directly from here click on draft and so on but yeah >> this is the way you could create a dashboard if you don't have much knowledge about how to create dashboards but you have strong uh domain knowledge >> and I have a question the dashboard that is currently published can we modify because when you want to publish it's

4:19either ambit credentials or it's a user permissions Which one is used for this dashboard? >> That's a good question. On our case, we're taking the embedded we're taking the on behalf of users. So, we're taking my uh credentials because that's a very good point. We don't you don't want that everybody uh have access to the to some sensitive data. You want to still keep

4:41the separation that you uh have set at the command table for each users uh of database. So, we use my credentials to get access to this data. Someone who don't have the same credential as me then might not have the same permission. It will not probably see the same type of data. So you will you will probably see some errors probably displayed or some difference of data displayed.

5:05>> Awesome. And in case they want to deploy the same apps, they just need to uh clone the ripple that you'll be sharing with us in the description of this video and then deploy Databricks apps. That's right. >> Exactly. The only thing they would need is just to if I go here, they would just need to have um a token uh and then uh the work the SQL warehouse ID that they

5:29will have to uh write in the in the code and then that's it. Everything is done before. They just need also to have the data ready obviously and then the better data the better result they will get and then yeah they can directly create as many method that they need to conduct amazing uh strategic operations. >> Awesome. Thank you so much Kristoff. So you can witness how the field is like

5:52creating some awesome tools using Databricks platform. Thank you Kristoff so much. >> Thank you.