Build No-Code AI Agents with Kasal (Databricks App)

0:04Hey folks, we are back for another video for the castle series and today I have the pleasure to host Anu Roy. Hi Anu. >> Hi. Hi Yousef, thanks. Thanks for having me here. A little bit about myself. I've been with Databricks for under a year as senior solutions engineer and it has been my pleasure to collaborate with Kazal team and finally be a part of the series to you know for the next step

0:28>> and as you may know I have a very bad memory so please can you remind me what we talked about during the first two videos >> of course that would be my pleasure so I think all of you here know about kazal if not maybe just a quick recap a kazal is basically a datab bricks app right where you can in no code fashion be able

0:50to deploy agents and orchestrate. So it's not just about agents, it's really about delivering use case. So if just to jog your mind a little bit here is the UI how it looks like on on uh the framework. On the left hand side it's really more about configuration how you configure different tools and on the right is really just a chatbot where you can get started and you would be able to quickly spin up uh you know agents that

1:14would be doing tasks for you. Does that jog your memory now, Yousef? >> Yeah. Yeah. Yeah. And and I and I do remember that we can deploy the Casal app straight from the marketplace uh or we can do it from the Casal repo in GitHub. >> Yes. So that's that's actually a very good point. So for really to people to get started super quickly, it's it's really easy to come to your Databricks

1:37environment, go to your marketplace version, search for Kazal. Our team is continuously working for more and more updates. So it's just a one-click install and it really deploys on your datab bricks environment itself. And as you see right now the uh the instance that I've hosted here is running on my workspace. But there's also uh an another option to uh you know if you want to stay updated with the latest

2:00features that the Kazal team is working on. There's of course the solution is open source. You should be able to install the from directly from the GitHub repository and be able to spin up on your local local as well. >> Crystal clear. And I think that was that's what we mentioned during the first video. But what about the second one? I think I think that's where we introduced like perplexity and jin if I'm not wrong.

2:22>> Yes, maybe here as well. Like I said, so on the left hand side it's really about you know configuring kazal how you would like what kind of tools you would like to play with with your agents and tasks. And I think last time what uh what my colleagues would have shown you would really be to go into Kazal and be able to uh configure a genie tool a plexity tool and today we are here to talk about yet more features that Kazal offers you.

2:46So uh today we can we will be talking about two new features let's say. So within Kazal it's it's a databicks app right and to be able to have one app for each person is not very uh efficient. So in Kazal today you are now able to create workspaces. So what essentially that means is you can uh create new workspaces and be able to collaborate

3:08within the same workspace. So right now if so this is an app where you see this I'm working on my workspace but just for a little bit context you should be able to add more people to your workspace be able to collaborate within within the same UI and this is really powerful because you can now be able to configure agents tools at a global level so it it's really one app on your production

3:30server where two three people from your marketing team or it could be uh really from the pricing team can collaborate on one instance >> and can you tell Okay. Like or tell me more about what we're going to talk about today. >> Yes. So today uh what I have in mind for you is a cool user story where you can think of it as a as a pricing agent,

3:52right? A pricing assistant who would be working on to uh to fetch the internal pricing of the company. So in this case I've taken an example of a toy company. And you can imagine there are there are different uh toy companies out there selling different kinds of products. uh there's a task given to a pricing analyst to be able to define what is the pricing strategy of the company whether they should increase the prices or

4:14decrease the prices. How would typically that happen is first the analyst would be able would be conducting market research right into market research going out there to understand the competitors and their prices and at the same time be able to you know correlate that prices with their internal data and finally be able to you know invest a lot of time uh investigating analyzing and

4:36then creating a slide deck for your uh yeah pricing strategy whether we should be placing higher prices or whether lower prices. Now, typically what that would take about a week. I think that's a little bit optimistic, but definitely two weeks of time. But with Kazal, what you're able to do is just quickly spin up that with uh just writing a prompt. So, here if I tell uh the prompt to, you

5:01know, create me a plan that positions pricing strategy based on internal and external data. And you see uh Kazal will I mean of course there's a lot of typos here but uh Kazal would actually start thinking and creating agents and plans automatically. In the interest of time what I would show you here is while it

5:23is spinning up I have already you know done this in the past and maybe just quickly showing you this would what the you know the agent would spin up and it would say okay typically there would be an internal data analyst fetching internal data. there would be somebody who's you know really collating data from the market and then finally be able to connect those dots together. Uh let's

5:45see if this is happening. Let me just uh execute it once again. Okay. And and so how I how I was able to connect to internal and external and that is the key piece of the information that I want to bring today with Kazal. So in datab bricks right now if you go to your agents section there's MCP servers available and with this MCP server you're also able to connect to Kazal. So

6:08what it essentially it means if I take a look at my DBSQL MCP server I have already configured in Kazal my uh MCP server to be able to connect to the DBSQL server and this is just a click of a button. So you basically configure the URL that you get from your workspace put put in your API keys which is really your uh authentication token and and

6:30that's it and once I start quering this data so here in this particular agent and the task planner what I'm doing is there are really three different agents like I said so one would be really handling the internal data one would be handling the market data and then finally be able to create that presentation together. So with a click of a button you see a new task spins up. It takes a little bit more time. And

6:53here if you see in my past results uh I have already been able to get some reports and here you can see that uh it it gives you a the entire details about how my u how my you know products were placed. So in this case it says that there there is an definitely an opportunity for price adjustment. So if if you look at my I can also show you

7:16the data but essentially it's a toy company. So it's it's basically having a lot of board games. It's also having traditional toys, electronic toys and there is positioning where you would be able to increase your uh prices for a fair competition. And you see here they give you recommendations as well. So all in all what happened was initially that would have taken about two weeks of time

7:39to collect all this data to be able to come up with a strategy with a click of a button. Now Kazal is really delivering results from two weeks to just 5 minutes. Uh and behind the scenes these are really LLM models that are hosted on your Databricks workspace. So you're making the best use of all the tools that's available already in your platform through this intuitive UI. Let's now you are also able to download

8:03this result. So here you see that uh based on different categories what are the current price trends and then they give you the u recommendation where you should probably be either increasing your prices or decreasing it or probably you know spend more time in the volume game. A lot of the time it's not about just increasing or decreasing the prices but rather focusing on the sales aspect

8:25where to position your volumes. And I think all of this is uh really very powerful if if you leave it to the hands of the business users. So that pricing analyst does not have to wait now two weeks to be able to show it to their leadership but it's already available in in 10 minutes. And in in here maybe also just a quick um the new feature that we would be also having a timeline just to

8:48see that how agents and tasks are progressing. So with the with the timeline you are able to see really minute by minute what was the uh what was the task that was conducted so what is the memory reads what are the LLM responses and you are able to see everything and I think now my crew is completed. So in in short term I would be also able to see a new result.

9:11>> Yes. It's it's backed [clears throat] by the MLflow tracing. >> Yes. Yes. You if you if you enable the tracing here uh also on your uh configuration it will actually all your tracing and evaluation would be logged into your databicks workspace. >> That's amazing. [clears throat] That's how you can replace [snorts] let's say the work that you used to do in maybe

9:33two days or whatever time required to maybe matter of few minutes. >> You would be surprised it actually takes more than weeks sometimes to compile that information. I just wanted to also highlight the powerfulness of Kazal here. So it's not really just reports but you can also build fancy slide decks to you know to collect that information so that there's the people spend more time on strategizing and making

9:55decisions on the data rather than really nitty-gritty and diving deep into collecting information. So kazal is really powerful when you combine uh your internal data set through MCP servers and also be able to connect it with with the information that influences uh from the external market data. So that's it. >> That was amazing. That was uh you're sharp. You were like crystal clear.

10:18Thank you so much, Anu. And maybe see you for another Kasal video. >> Of course, I would have be happy to join and stay tuned for more Kazal updates. There's a lot going on and the team is really happy to continue this series. >> Thank you, Anu. Bye-bye. >> Thank you. [music] >> Data seas lake flow streams the flow. Iceberg and delta lake tables of glow.

10:40Unity catalog governs league base oy [music] text to SQL MOSA AND KA PROLONG you've been quitting co-host light the fire dream scale higher [music] and higher next drop