Building Trusted Analytics with Genie & Metric Views on Databricks | Genie Demo

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Building Trusted Analytics with Genie & Metric Views on Databricks | Genie Demo · Channel: Databricks · 15:07 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Data, databases and embeddings, Privacy and data protection

Summary

The demo shows how Genie combines business data with organizational knowledge to answer broad questions about a healthcare business. It uses ontology snippets, dashboards, Genie agents, and SQL analysis to investigate vaccine performance and identify issues in Germany, including site readiness and payment authorization delays. Genie then searches connected systems, finding no relevant Gmail field notes but locating a risk report and recovery plan in Google Drive. The demo also shows how users can inspect sources and schedule follow-up updates.

Key points

Genie can use organizational knowledge from sources such as email, Drive, SharePoint, and Confluence alongside data.
Ontology snippets help Genie interpret broad questions and connect relevant business assets and knowledge.
Users can inspect snippets, related domains, people, and assets to understand where information comes from.
The Discover page presents curated business domains, including dashboards, Genie agents, metric views, and tables.
Genie investigates vaccine performance in Q3 and provides a detailed analysis of the injectable portfolio.
The Germany analysis identifies site readiness, conversion, and payment authorization delays as issues.
Genie can search external systems through configured connectors, using the user’s own credentials.
A Google Drive risk report and recovery plan provides operational context that is not shown in dashboards.
Users can schedule an analysis to receive updates on a report over a specified period.

Chapters

Start	Chapter	Summary
00:00	Genie and organizational knowledge	The presenter introduces how Genie can combine data with knowledge from across an organization to answer questions.
01:03	Investigating vaccine performance	Genie uses business context and ontology results to analyze the healthcare business and its injectable portfolio.
04:43	Exploring domains and business assets	The presenter explores curated domains and assets, then reviews Genie’s deeper analysis of performance in Germany.
07:49	Germany analysis and root causes	Genie identifies issues including site readiness, conversion, and payment authorization delays, with sources for its findings.
09:22	Searching connected systems	Genie searches Gmail and Google Drive through configured connectors to find field notes or a resolution plan.
12:29	Finding the recovery plan and scheduling updates	Genie finds a Germany risk report in Drive, and the presenter demonstrates scheduling future updates on the analysis.
14:03	Scheduling follow-up	The presenter explains that Genie can send scheduled updates so users can monitor the issue without repeatedly checking it.

Quotes

Genie can understand that information, infer knowledge from it, and get you to an answer quicker. — 00:31

just by having that ontology layer ready and at hand, it can quite quickly pinpoint the direction of this conversation — 03:40

I can see the sources — 08:21

it really has the operational why behind the numbers that the dashboards don’t really show. — 12:59

I didn’t have to ping anybody. Resources were already available to me. — 15:07

Concepts and entities

Genie, Databricks, ontology snippets, domains, Genie agents, Metric Views, MCP, Google Drive, Gmail, Unity Catalog

Full transcript (original language)

00:00 [music] >> Um, we’d like to highlight um, how you can get a bit more confident in your answers. So, how you can interact uh, with Genie 1 based on not just data, but also organizational knowledge, whether somebody sends an email or somebody writes a document. Um, people still have a lot of meaning that lives in their heads. Um, and Genie doesn’t just shorten your your way from, “Hey, I have a question. I’m going to generate the query and this query will work.” No, it

00:31 really allows you to go way beyond that to start reasoning across the whole data ecosystem and not just that, but also um, across the cumulative knowledge in your organization, whether that’s in emails, in Drive, in SharePoint, in Confluence, anywhere else. Genie can understand that information, infer knowledge from it, and get you to an answer uh, quicker. So, let’s uh, let’s have a look. So, let’s get to a quick demo and briefly talk about how quickly you can

01:03 get from a pretty broad question to leveraging the knowledge of your whole organization without having to go through 10 different teams um, to get the information that you need. So, right now we are in Genie 1, which is our main gateway to interacting with Data Lakes. Um, we are interested in our healthcare business. We just closed Q3, and I’d really like to understand how how this went, and I’m hoping that Genie can can help me. Um, within the business we deal with two sides. We deal with vaccines, so

01:34 injectables, and we also deal with oral medication. Specifically here, I’d like to understand more about our vaccines business. Um, as soon as I ask this question, Genie’s trying to understand and narrow down scope a little bit. Um, and you can see the full detail of what Genie’s doing. It’s looking at some past conversation, but also at other relevant assets in in business context. It’s looking through a bunch of different dashboards and Genie agents and different tables which might have related content, but it also found directly nine ontology results for

02:06 things related to our vaccine business and performance and metrics in Q3 etc. And we can see what these knowledge snippets are. They’re automatically inferred by by the platform across the variety of assets that you currently have. If you want to understand more about some of these snippets and how they are defined, you can simply click on them, kind of navigate, understand more about how they’ve been created as well as various different elements that they relate to. So, we can see that this this particular snippet is

02:38 related to a few different domains. Domains really allow you to organize data and AI assets around how your particular business is structured. You can also see who are the relevant people who are related in some way or another to this particular snippet. And also any other assets that that might be a help for you. And you can navigate all of the different Folk and assets and snippets that are related to each other and you can click and and navigate through all of that to

03:09 understand a bit more if by any chance you need to to look in other sources if if Genie doesn’t do that for you or alternatively you want to guide Genie in a particular direction. So, we’re we’re going through all of these results as um as Genie’s trying to come up with the answer. So, look through the snippets. It has some information from like prior conversations that we’ve had around our our injectable portfolio in this particular business. It asked a few

03:40 specific questions to various Genie agents. Quite targeted question because we also have built in specific Genie agents for for some of these purposes and it’s getting some of those results and curating those for us. Um has quite a bit of confidence in the data itself. It’s executing its own additional uh additional queries and generating some visualizations. So, overall, just by having that ontology layer ready and at hand, it can quite quickly pinpoint the direction of this conversation, which is

04:12 immensely useful if you have a pretty broad question about your business. Um and you’d like to to narrow that down, especially when it comes to uh large businesses with various different cross-functional teams. So, Genie thought through all of that. We can see the whole thought process over here. And we can see um now a pretty detailed report about our injectable portfolio, what the performance is, how it’s been comparing in the various different markets, um some different results. We can see that there’s some issues in Spain and

04:43 Germany. Um so, overall, it gives me a pretty good uh pretty good view of what’s happening here. Um so, as I kind of go through this, I see that there’s like some some issues in Spain. Uh but at the same time, there’s um also some in Germany. Um I can pick up that site readiness is usually a problem that we might have had before, and we know that this is kind of affecting our um our business quite seriously whenever this is uh this is happening. So, we’d love to to learn a bit more.

05:13 We can follow up with Genie to understand what specifically is happening in Germany. Um Genie will go and do a a further detailed analysis um across the Genie agents, having various different queries, etc., and um and do that analysis for me. In the meantime, earlier on I mentioned domains. Um it’s something I want to I want to highlight because it will be a key part of not just your journey over the next couple of days, but more specifically once you start producing some of these dashboards and Genie agents, it’s one of the ways in which

05:44 you can start engaging your business and telling them more about the the data and assets that are at hand. So, if we go into our discover page where we can see the various domains available in this workspace. Um like I said, you can explore um these and you get to define these by different by business area. You can see we have quite a few here in this particular workspace that we can browse through. We can also apply different levels of filters in terms of certifications and see what certified assets are, what popular assets are, etc. Um we can see

06:15 that there’s already a couple of domains related to our healthcare business. Um there’s something here available for um commercial and and uptake for for some of our uh vaccines. Um and this is basically a curated view that shows the assets that we want to put in front of uh of a business user. We can see a few different dashboards um or Genie agents if they’re available here. We can see whether they’re certified or not. Um so, meaning that somebody has defined that this is a curated asset and should be um

06:47 leveraged more than um than assets that somebody might create just because they want to do a one-off analysis. We can We can see information around metric views and various tables and a whole bunch of different statistics around these. Um so, very quick to see the information that’s available related to one particular part of the business and you can help customize this um according to how your business works. And based on some of this metadata, Genie will also take some of that and um self-infer ontology snippets which will further

07:19 help as you are asking um some of these broad questions to um in your Genie conversations. So, if we go back to our chat right now, we can see that um Genie went through um and ask our particular exact analysis Genie space a few different questions. It dug into a bit more detail there. Then it realized it needed to do a bit more analysis by itself. So, then it went and ran its own its own SQL query, and now it’s just um generating some some final charts for me um at um

07:49 um at this point. Um now, if at any point I’m um I’d like to to get that answer quicker, I can always skip. Um it just means uh Genie will try and terminate some of the some of the running queries and get me a summarized answer. Um this did happen to finish quite quickly. It gives me a pretty detailed deep dive of of what’s happening in Germany right now. Um kind of tells me that um the injectable portfolio is a problem. And as we go through some of this analysis, we see that it’s some very particular drugs. Um

08:21 and more importantly, the root causes for this. We can see that the sites where these vaccines are administered, they’re not really ready to administer the vaccines compared to some of the some of the other markets. And for every one of these answers, by the way, I can see the sources, right? There’s specific ontology snippets related to this with contextual information and specific queries that point me to the root cause of this. So, site readiness is one issue in this case. And another one is the the conversion rate in general for oral to

08:51 injectable, since we are trying indeed to move folk to our injectable medication, since it’s it’s usually much better than our orals. And then there’s also some authorization delays in terms of in terms of payments, right? So, there’s a few different issues in this market. Um and Genie already has some ideas of of what it of what the a resolution could be. But in general, data usually tells only a part of the story, right? We’d like to understand if anybody has already done some of this analysis that Genie might

09:22 have not picked up, because maybe it’s not in this environment. Let’s see if um we can leverage some of the existing connectors. So, Genie can also uh use MCP servers and connections into other systems. Um also, you can leverage pre-built skills that you might have in Genie, um, to to help further your analysis. In this case, I want to understand if, um, there are any field notes on Germany from anybody on my team or any other team who might have shared an email to, uh, to one of our Google Groups. So, as we

09:52 start, um, asking Genie whether that is happening, um, it’s able to call some of these, um, external tools for various searches. So, you can see here again, you can see a lot of the the trace behind this, where, um, Genie’s going and searching Gmail for some very specific things, again, based on the inferred context of the conversation and the ontology snippets that it has. Um, so, right now it’s it’s doing, um, it’s doing a bunch of, uh, of that search for me and trying to understand if there is, um, anything,

10:24 um, in there that might be helpful for for this. Cool. So, as, again, as Genie is loading, um, the we would, um, we would sometimes try to nudge it in particular ways. Um, sometimes it might find a result, sometimes it might not. Um, particularly in this case, it did not find any, um, any field notes, um, on this. There’s nothing related to

10:56 to our particular medication. So, we could try and give it a nudge to see if, um, it it found something else, or we can start looking at, um, at some other sources. Let’s see if it can actually find something in Google Drive instead. Um, so, if we start thinking about what could be there in Google Drive, um, maybe some, any other field notes, uh, or maybe a resolution plan for the issue in Germany. Um, let’s see if it if it can find

11:27 anything there. So, again, you have quite a bit of freedom and flexibility, um, in terms of where Genie can look, especially if these connectors are set up within your workspace. So, there’s a wide variety of connectors I have two already, well, I have uh, more than two already set up by the administrators in this workspace for myself. I have connected with my own credentials to these two. I can also connect to some of the others. These are basically just an, um, an OAuth, uh, prompt, um, to to make sure that I have

11:58 the the relevant authentication to these systems and my admin my admins can set up more connectivity for me. This makes sure that everything that you do access in your conversation with Genie is something is data that you actually have access to, um, and you’ve been given you’ve been granted access either through Unity Catalog or through, um, any of your other systems since everything is using your, um, your own credentials. Now, um, we’ve not managed to find anything in,

12:29 um, in Gmail, but we did find something in Google Drive, right? So, again, Genie went and tried to understand a bit more about is there anything related to to Germany and um, and any of our, um, ontology snippets, uh, that we’ve we’ve kind of, um, discovered in the conversation and it did manage to find a few different documents in, um, in Google Drive. Um, it found a Germany risk, uh, report and recovery plan, um, which is, um, updated

12:59 recently. It’s prepared for a board risk review and it really has the operational why behind the numbers that the dashboards don’t really show. Um, it’s information that, uh, some of those teams have put together because they know the market much better than we do. They didn’t happen to share it over email this time, but they do have it in, um, in Google Drive somehow somewhere. And we can always go to the root source of this. So, we can see that, hey, somewhere in Drive there’s, um, there’s this risk reports that somebody has prepared um and there’s a full picture

13:31 um of what’s happening with our payment system and site readiness gaps, etc. So, Genie managed to get from a very, very high-level question to something quite specific that is already being action and it gives me quite a bit of confidence that the local teams have this handled and I don’t really need to do uh much else in order to um to follow up on this issue. However, if I ever did want to do that, if I scroll uh if I scroll um to the bottom of this to understand um

14:03 what else it had found. By the way, it shows me all the different sources. We can see that there’s been a few different other reports um here on this uh on this particular issue. If we want to repeat some of this analysis, we can always schedule a task um straight from uh straight from Genie, we can ask it to schedule this analysis. Um we can even be more specific if we want to. So, for example, if I want to stop this, I can say, “Hey, schedule this analysis for me specifically to give me updates on the on the report uh for the next 2 weeks.”

14:35 And ultimately, what Genie will do is it will create a schedule for me um where um I will be informed via email or any other sort of preferred connectivity um on a particular time and date um about the updates on on some of these reports, which is fantastic, right? Because I don’t have to monitor anymore. I know there’s a I know there’s a problem. Uh folk can um can solve the problem by themselves um and I can just keep an eye out to see if uh if there’s any issues. So, we

15:07 managed to get there quite quickly, quite simply. Um I didn’t have to ping anybody. Resources were already available to me. So, I managed to make the most out of um of what was there. >> [music]