Building Intelligent Event Discovery Agents with Databricks Genie

0:04Hey guys, welcome back to Casal series. As you remember in the previous video, Neto, the co-founder of Casal introduced the reason why he built Casal and how it makes the Asian deployment much more easy. I have two guests, David from datab bricks and have max from the customer side who will talk about the uh

0:28issues they faced and why casal was the right solution for them and then David will introduce two new components and just keep in mind this is a series of videos so within which each video we're going to introduce additional features so at the end you will be able to build an end toend agent on datab bricks easily. Hi Max. Hi Yousef, nice to be

0:50here. So I work at Coca-Cola Hellenic Beverage Company. I am in the digital technology and system platforms team and there we work on data strategy and governance. We also do innovative things with regards the strategic part including the thing that we'll present today. We had business users coming to us from one specific business unit. The

1:14idea was you know we're a big beverage company. We tried to increase our revenue by selling more of our product and for one specific brand in one of our business units sales and marketing team came to us and wanted to have sort of solution that can find events that could be relevant for that given brand strategy. So if they say you know we

1:37have a certain brand we want to target the goal would be to increase distribution how can we do that if we had a look at very specific events. So within given target maybe locations or hubs uh can we somehow do that because what does this usually mean if we were to tackle this? It means we need somebody from the event team for example that will do the research on which

2:00events have we participated in before maybe with other brands um or just generally which ones are available because some of them might be reoccurring but there might be new ones as well for next year right uh but then not of all of them would be relevant so we need to curate them as well event and which ones are located in the cities and hubs that we're interested in then we have the coordination between the

2:22different departments because those are just the events We need maybe the marketeers to identify is that something that we can do? What exactly can we do there? What is the action proposal for this specific event? Um and also the whole categorization and relevance assessment. So is this even relevant at all for the brand has to happen as well. So we have the different stakeholders

2:44from commercial excellence port or portfolio team uh the event team and the marketing tier all coming together aligning meeting up. So what we tried with Cassal is basically including those key stakeholders that we would usually have in the setup in the as a multi- aent system and trying to automate the

3:06event discovery to automate the relevant scoring to automate the coordination between who receives which input. Uh making some actionable proposals depending on the strategy as well. With that being said, we had implemented as mentioned the overall framework uh which David will present in detail.

3:30Thank you Max. My name is David. I work for Databricks since roughly half a year. Um I'm solution architect for Coca-Cola and had the lovely chance to work with Max especially and Nema together on the use case which we will show you in a second. So um as Max briefly introduced in order to solve this complex real world challenge um let's probably start directly first from

3:52the business side right in order to solve it like you have might have hinted from Max you would need in a real world setting something like an analyst a portfolio manager somebody who does web research or a marketeer right and what we now try to do and what I will show you now is how we solve this challenge together with our customer on a real world business use his example. So, this

4:14is a bit of a more complex crew as you can already tell, but I will guide you through um every little piece of the way. Okay? And also probably important, we even added one little component based on on reality. Imagine for instance we have a setting where we have some sales numbers stored in a database and business user might come on top of this strategic marketing mapping which maps

4:36explained um to position certain products that they also might to want to run this process when they see declining sales numbers which usually you see like a pattern in your data. Okay. And so basically what we will have a look at and let me first also run it while we talk we can already um have a look at the results once we are finished. In order to solve this complex business

4:58challenge in a real world scenario what you might need is for instance you might have some kind of analyst who does some kind of web research. You might have a web researcher portfolio manager or a marketeteer in order to solve the real world challenge which Max explained. And what I now prepared is I basically put one um item in front of it. This is let's imagine you don't want to do a

5:21general web search for a certain product but you come from this business perspective that one of your products let's call it for instance product A um that it has declining numbers in sale and you now want to find certain targets certain event locations where you can position the target based on your brand strategy. So this part which Max explained in order to solve this challenge we have here actually a quite

5:45complex crew um but no worries I will guide you through every little piece of it. So the first technology in order to have a look at the data of what kind of product is declining would be a genie space from a data per uh bricks perspective. So I just present here basically dummy data. So what this means I created some random data where you have happen to have on product A uh a

6:07reduction of um of your sales and what happens now in the background is that this genie space I cultivated one question so which in which city do we have for this product a um a reduction in sales by x% and what now happens is literally the parallel stream because you now have on the one hand your brand manager and your strategic manager who

6:31knows so much about your business. So your strategic priorities, what kind of target groups you want to have for for um this certain product and in parallel you also have the web researcher who finds basically who does internet research and finds web information about which kind of events um happen and what now happens in parallel those two

6:53streams they are processed and ultimately they are fed together to a marketing manager. He basically gets the input from your strategic manager and from your events to find basically mathematically speaking the intersection of where to best put your money in order to solve your original question which you had that declining number in sales which you had. Okay. And ultimately we

7:16just have some nice formatter to basically demonstrate um that this is nicely formatted into a business view of Excel table like structure. to solve this on a technical level. I already ran this crew. What we had to do in the background very easy. Um my colleague Nema mentioned it the last time on the datab bricks landscape. We will now um introduce one after another the

7:39different tools which we have in Kasal. And for this little workflow we actually use Genie spaces and for web research perplexity but there's a whole v variety of other tools which can be used. And actually the final output is then just you design your workflow either by natural language or directly you can create the agents yourself. And at some point do you see this is still running

8:01because it needs around 5 minutes. You ultimately see the nicely formatted output where in my dummy data I basically had in the city of Bologna a declining uh sales number and it found literally all the most important information about a certain event and literally which product to target. And this is ultimately how the entire business challenge most solved for Coca-Cola not on dummy data for them

8:24obviously on their own real world data. >> That's really amazing. And and can you show us like quickly how we maybe we can integrate uh Genie? >> More than welcome Yousef to also show you that. So it's super easy on the Casal side. So you have on the top left side this configuration um button as my colleague Ne briefly mentioned the last time we have something like a workspace

8:47concept. So every tool which I show you, you will um allow it on a workspace level. So in this case, we're in my home workspace. And allowing Genie is as simple as you have here your list of tools. You simply can add them. Let's add Genie. Let's add Perplexity. You need to then allow them. It would hint you, for instance, because for Perplexity, you also need an API key. Um

9:11it would then hint some kind of message. Hey, you forgot to set an API key. I will show you in a second where to set that one. And we allow Genie for that. You don't need an API key. And then if you have some kind of API key missing error message, you simply go here, set your keys. You will see this for me. Perplexity is already set. And this is it. And you're good to go and can use

9:33the tools out of the box. This is as simple as it can get. >> Yeah, that's so simple. Doesn't require any coding knowledge. It's >> exactly UI based and that's it. So you can really focus on your business challenge and how to integrate this in a workflow. >> You know how to integrate perplexity and jinny. Stay tuned for more uh features.