Self-Service Analytics for Business Leaders: Comcast's Genie-Powered Dashboard Platform

Summary

Comcast's commerce platform engineering team solved the 'one-more-chart' bottleneck by building a self-service analytics platform using Databricks Genie, AgentBricks, and AI/BI Dashboards that lets senior leadership query and modify dashboards in hours instead of weeks.
A 30-60-90 day implementation plan established medallion-layer pipelines and certified metrics views by day 30, dashboard automation with Genie Code by day 60, and fully automated feedback-to-production workflows by day 90.
Spec-driven development with AI code generation and embedded feedback loops convert leadership requests into production-ready pipeline changes in minutes, while Unity Catalog certification ensures KPI governance and stakeholder trust.

Self-Service Analytics for Business Leaders: Comcast's Genie-Powered Dashboard Platform

Comcast's Senior Leadership Team faced a critical bottleneck: business leaders needed fast analytics insights but remained dependent on IT and analyst teams. Standard report requests took weeks, and iterative changes meant longer delays. Genie, AgentBricks, and AI/BI Dashboards unified data management across multiple tools into a single analytics platform built on Databricks Unity Catalog, enabling self-service querying and dynamic dashboard generation in hours instead of days.
This case study shows how Comcast's engineering team transformed their data architecture using medallion-layer pipelines, spec-driven development with AI code generation, and embedded feedback loops. The result: leadership teams can now ask natural-language questions, slice data by any dimension, and request dashboard modifications without waiting for analyst hand-offs. Automated processes convert feedback into production-ready changes in minutes, reducing development cycles while building trust through certified KPIs and governed data.

Chapters

FAQs

What is the one-more-chart phenomenon in dashboard design?

The one-more-chart phenomenon describes the cycle where every completed dashboard triggers a new request for an additional dimension, segment, or chart, requiring another round of analyst involvement and engineering work. Comcast identified this as a core bottleneck because each iteration created a delay loop that often outlasted the original business need.

How does Comcast use Genie for senior leadership self-service analytics?

Comcast embedded Genie into a Commerce Insights application that lets senior leadership ask natural-language questions about sales and customer experience data, slice results by any dimension, and request dashboard modifications without submitting a ticket. Certified KPIs governed by Unity Catalog ensure that Genie's answers are consistent and trusted across the leadership team.

What is spec-driven development with AI code generation?

Spec-driven development means writing a structured specification of what a dashboard or pipeline should do, then using AI code generation tools like Genie Code to produce the implementation from that spec automatically. Comcast uses this approach to convert leadership feedback into production-ready Databricks pipelines and dashboard changes in minutes rather than days.

What is the 30-60-90 day plan Comcast used for its analytics transformation?

In the first 30 days, Comcast established medallion-layer pipelines and certified metrics views in production on Databricks. By day 60, the team deployed dashboard automation using Genie Code so natural-language requests could generate updates automatically. By day 90, a fully automated feedback loop converted user-submitted requests directly into production pipeline changes, collapsing the development cycle from weeks to minutes.

Full transcript

[00:08] Thanks everyone for joining today. Normally when we schedule this kind of sessions, afternoon is kind of a laid-back, less crowd. But thank you all for joining this today's session. Uh you will have a good time here. So, my name is Verghese. Uh I have joined by Balaji and Satya
[00:25] from Comcast. So, what we want to kind of walk through the journey that we went through for the last year and a half 2 years for the AI transformation. So, what we do as a team is basically look at the
[00:41] customer experience, customer friction points from a sales and uh order perspective. So, that's what our primary focus is. So, what I want to set the context is that that will help you all to relate to us what we are struggling with. Last year,
[00:57] one of the biggest struggle that we had was that business want to know everything so quick as possible. Business wanted everything yesterday. And they don't know how to ask and give you the requirements. If they can give us a good requirement, we should be able to give the right data
[01:14] points for the our consumers to do it. So, but that is what we are going to talk about it. And Balaji and Satya will walk through in their deep demo session. So, that's something that we want to go through. So, as you go through a customer journey or
[01:30] any requirements, I'm pretty sure most of you here been approached by somebody asking us, "Hey, can I get this report?" And that's the question you all kind of face every day. There's no requirement. There's no clarity. Nothing is there.
[01:46] So, what are you going to do? You're going to take that requirement. The analyst will run in and get the data and saying, "I need to get this data from this table, that table." And you assume a lot of things, and then you're going to do work with their data engineers to
[02:02] build some pipelines and give it to the build a report and sit with the requester. Guess what? Requester is half happy. Like, oh, can you slice this data differently? Can you do a dimension differently? Can you segment this data?
[02:17] So, it's just going back and forth. It takes a long tedious process from a requirement perspective and all the way to the execution. By the time you get the report ready, the team the person who is asking the requester the requester is gone. They've
[02:34] moved on to the next project or the next initiative. And that's what happens. We all have the pain going through the process. So, beginning of 2025, what we did was that how do we solve this problem because our senior leadership is like, we need data,
[02:50] data, data, data, but we couldn't keep up the speed and the demand. We all talk about data democratization. Constantly you heard over and over again in the sessions. How can we get to that point? So, we kind of kind of brainstorm and said, okay, we did some POCs
[03:06] and a lot of things that we tried to do some AI AI models and Genies different versions. Then all of a sudden, last year conference, Balaji and Satya came back and said, again, hey, guess what? Databricks is coming out with the Genie.
[03:22] And that is where kind of helped our transformation kind of speed up their transformation. So, what we did was that look at the Unity Catalog, data pipeline, Genie, everything in one place, build with their dashboards that is already there.
[03:38] So, with that with that enablement, our transformation from went from a one year one and a half year journey to a six months journey. And I think team has worked a tremendous job to get to where we want to do it and that's kind of help
[03:54] you to see that how quickly our end users can go and ask the genie, can you slice this data anymore? Can you slice the Can you convert this into a pie chart to a line chart? They don't have to come to the dev team or analyst to do that. Can you give me these Can you segment this data into a region or
[04:11] market? They can ask that question themselves rather than depending on an engineer or a analyst to do that. So, that is a journey I want to open up this conversation with the team and they will walk through step by step and how we got there. So, it's a good journey, sit back
[04:27] and enjoy because you all go through the same pain that we went through it and hope this going to help you to get to the transformation quicker. Thank you. Balaji.
[04:43] Is this all right? Sorry about that. All right. So, I'm Balaji. I'm the senior manager for the data engineering platform at Comcast. I have Satya with me. He is my principal engineer. Just to set a little bit context as to who we are,
[04:59] right? So, we are the back end engineering team for the Xfinity commerce platform. Um we have data all the way coming from the sales platform, right? Customer browsing the products that they wanted to
[05:14] purchase, you know, once they purchase, then it becomes an order, then order that goes through a several workflow until customer, you know, activates the device, right? So, the data is being generated all the way from the sales platform and also
[05:31] from the ordering platform, right? And the data we collect, we ingest the data for our, you know, reporting and analytics. A typical data engineering team, right? But then, um there are three areas that we focus,
[05:47] you know, um one is operational efficacy, right? Every order that we collect should be submitted and then it should be closed, right? So, we make sure everything that we get today, um are fulfilled and closed, right? That's
[06:03] our intent. Anything that is falling out, anything that is delayed is when our monitoring and reporting kicks in and then we engage our right teams to um you know, look at those issues. So, that's one of the operational efficacy.
[06:19] Then the project launch effectiveness. You know, at Xfinity you know, we kind of launch a feature every 2 weeks, every 4 weeks. There is a feature that goes live, right? So, we would want to measure the effectiveness of all these
[06:35] features that we deploy, right? We We look at what what was the you know, the ordering rate before the launch and the ordering rate after the launch. If there are any uh you know, spike or a dip, we investigate that. So, that's the project
[06:51] launch effectiveness. Then um your typical you know, um um area where you, you know, look at a particular domain, go in detail and investigate further, right? So, that's
[07:06] another, um you know, unraveling business bottleneck where, you know, you take one domain, ordering domain, look at the trends for it and then if there are any, you know, um difference in the way the orders are flowing through, so that team focuses on
[07:22] it. So, that's the three areas that we focus on, right? It's typical data engineering team, you know, we don't have, uh at least within our hour, we don't have the, you know, BI expertise, right? Like there is a separate BI team, enterprise BI team who
[07:38] manage all the uh SLT dashboards, right? So, that's a completely different team, but then within our commerce platform, we have our own engineering team who also build operational dashboards, you know, for our own leadership and also
[07:55] other, you know, um, reliability engineering operations team, you know, a little bit of product team and little bit of business team also. So, uh, that's typical. And that's where we are. Just to set the context as to who we are. Right?
[08:16] What's our current landscape and the challenges, right? Um, challenges, I will, you know, uh, put it in two perspective, right? One is from the leaders perspective, um, and one from a engineering perspective, right? From a leaders perspective, if
[08:32] you see, um, every time they come and ask, "Hey, why is this, you know, a report building takes weeks, right? Why do I Why do you take couple of weeks not to build a report?" A brand new report, at least it's okay to take 2 weeks time, right?
[08:48] But then for a change that is coming after a report is built, a small little change, right? That also takes 2-3 days for them, right? And then there are quite a few dashboards, you know, even for one metric, you know, at least in in in our
[09:04] area, we do have multiple different, uh, dashboards available, you know, so there are people who get confused as to, you know, why there are so many dashboards, you know, um, why the same metric is being repeated in multiple dashboards, right? And there are multiple BI tools also, you know,
[09:22] that we maintain, you know, same metric is metric is being repeated in multiple different different BI tools. That's the challenge from the leader perspective. And from a engineer perspective, right? They feel that, you know, why is this,
[09:39] you know, a requirement are not given at one one stretch, right? Why does it have to come back with feedbacks many a times, right? Um why can't they give the requirement, you know, at one go?
[09:54] And and why do I have to build a report in multiple platforms, right? A same report is being built in, you know, um in multiple BI platform that we have. And the numbers are also different, you know, the data stores uh and the report, the way we calculate
[10:11] and show up in the uh reporting tools, they are all different, right? So, the So, the engineers and the leaders get confused, you know, as to why these numbers are different. And we always spend time and investigate and then, you know, make sure the number
[10:26] is correct and then go back and hey, use this report, this is what you should be using and things like that, right? So, there's always a challenge with the way we uh have in our current landscape. Um with that, you know, I'll let Sathya explain
[10:43] the challenges that we went through, right? Hey, you all. Can you hear me okay? All right. So, now you saw all the managers in the suits talking. Like, I want to give um an engineering perspective of how um how it really
[10:59] impacts on, you know, like executing all this, right? Like Balaji stated, like we come from an uh data engineering perspective, we our team may not have those BI expertise. Like, I'm going to talk through like top three problems along with others, which is
[11:15] going to uh which is like impacting us or which is like affecting us in delivering um faster timeline insights, right? So, I'll start with the number one problem here, which is um the one more chart problem. Uh you all would have gone through um
[11:30] uh this experience where uh you've been asked to build uh a a dashboard and then you start with a really neat clean version with minimal golden KPIs which kind of conveys the story and then you give it to the consumers, right? Now, one
[11:45] consumer comes in and then says, "Hey, can you slice this KPI by a dimension?" Right, for example, let's say sales sales numbers. Can you slice it by region or product, right? And then okay, yeah, we go and then add that chart in there. Now, the user two comes in and
[12:01] says, "Yeah, now can you add it as a timeline chart?" Right, I want to see the timeline in there. And number three user comes in, "Can you add two more KPIs?" Like you keep on accommodating all these you know, requests that comes in which of course we want to support
[12:17] our consumers to have access to all those data points. But what ends up happening is you you create a dashboard which has too many KPIs in there and it doesn't convey a story at all. It creates a cognitive overload for the users who go go through the dashboards.
[12:33] Now, what happens eventually is the users may not be like like it is right there though those KPIs are right there. They they have to scroll down through the dashboards to look for them, but since they're not able to like readily see them, they again submit requests for like, "Hey, can you add this one?" And from my
[12:49] perspective, hey, it's already there. That's that's the number one problem, right? Um number two problem this is this is going to resonate with many engineering teams right out there. Um this is what I'm I'd like to frame as an art versus engineering para- paradox. Like as an
[13:06] engineer, my mind works around logic. Um if I'm given a spec, I would like to you know, convert that into test-driven development and then I have I can write pseudo code and then code and then test it out and then I can call it a day and
[13:21] then like pack my bags and go home, right? Um that's how my engineering mind works. There is a closure to um what I've been like developing. But telling a story with data, um that's an art. That that takes a different kind of skill set. And there
[13:38] is a lot of conversation going on with the consumer. And as a data storyteller, and that kind of you know, lags within many engineering teams. So, if you are an engineering team and you've been asked to do this kind of you know, data storytelling, that is a challenge in
[13:55] you know, current landscape, right? That I'm going to go to the number three problem. Many um many BI teams and engineering teams have like you know, over the course of evolution have collected more BI tools. You would have started with tool A and then you know,
[14:11] like tool B came in and then some users are like good with tool B and then you build tool C which has you know, like more features and whatnot, right? Now, you have the same dashboards which go into um like multiple BI tools,
[14:27] right? And there are multiple consumers. And these leaders they talk to each other and then say, "Hey, the numbers that you show me here doesn't match with the number you show me there." And that's because our pipelines that we have built for each of these BI tools are are
[14:43] separate and they are redundant and they are like maintained by different teams, right? For example, let's look at this example right there, right? So, regional sales, it's the same same KPI, but because of the complexity of how data flows into different tools,
[14:59] the same number shows up in as different in three different dashboard tools, right? Now, as an engineering team, it creates me like two problems. One is how do I build this trust with my consumers because the numbers doesn't add up, it doesn't match. Number two is
[15:16] if one of these pipelines break or if I have to develop a new feature, I have to go through and then do redundant pipelines across, which means there are multiple points of failure. So, that is my uh problem number three, right? So, these are the top three um you know,
[15:31] like areas which we would like to um you know, like common um BI problems that you see around. So, I'd like to like pass it again to Bawaji who will walk you through our vision and how did we take it over?
[15:53] So, see we went back to our drawing room, right? Like we wanted to solve our problems, right? Like we wanted to do three things. Of course, we wanted to enable the chat um you know, like Genie-like a chat for our users to you know, to do the natural
[16:08] querying, right? And then we wanted to do uh a faster development, right? Like we do want to deliver things pretty quickly in hours if not days or weeks, right? And the third problem that we wanted to solve is
[16:25] is to have one um you know, a report or one source of truth instead of having multiple BI platform, you know, um have one you know, a certified truth for the all the data that we present in our
[16:40] dashboards, right? So, what we did um we did um a planning, right? Like our execution plan, right? With a 7-day, 30-day,
[16:55] 60-day, and a 90-day plan, right? With first 7 days, we evaluated the Data Bricks um you know, BI dashboard plus Genie, right? We took the data manually, right? Everything manual uh for the
[17:11] first 7 days. We wanted to check, you know, whether it will work for us. Right? So, took the data in an Excel, manually loaded that into Unity catalog, right? And then used built the dashboard and and had the Genie enabled. Right?
[17:30] So, that worked for us, right? Uh we did show, you know, to few uh stakeholders, everybody delighted, you know, wanted to have that, right? So, then we had to take a next step, right? In the 30-day plan. So, what we did, we
[17:47] took one domain, right? Like we have multiple domains in our data set, like one domain, right? Like um and we had to replumb the data, right? Earlier, we were did using Databricks. Databricks was more like a a Spark engine, right?
[18:02] We were not using Databricks for uh reporting, not for any other Genie uh products, right? It was pure Spark engine, no data was put in in the Databricks itself. So, we had to replumb all our engineering pipeline,
[18:18] right? And we had to follow the medallion architecture, taking the data all the way to, you know, from bronze, silver to gold. Right? So, all that we did within the first 30 days, making sure, you know, uh data is going to Unity catalog in a
[18:34] standardized way, right? Following the best practices. And we implemented the metrics view. Then we enabled the Genie, right? So, with the 30 days, you know, our one of our domain is ready, it's productionalized and rolled it over to
[18:50] our stakeholders for them to um you know, use it. So, that was another success. Then in the 60-day, right? We do want to automate the uh dashboard development, right? We tried We actually used Genie code,
[19:06] right? And then we um you know, we automated uh, you know, made sure we automated some of the dashboards, um, and and, you know, in the 60-day window, we were able to get most of our user
[19:24] community using the data for one of the domain. Then, in the 90-day plan, you know, we just finished, you know, not long ago, right? Um, in the 90-day plan, we we wanted to get a new app built, right? We wanted to
[19:41] embed the, um, dashboard and Genie within the app. And in the app, you know, uh, we wanted to take feedback from the users. You know, once a dashboard is shown, right? Um,
[19:56] users always have some questions, right? Hey, can you change this? You know, can you change this bar chart to a, you know, um, you know, line chart, right? So, we wanted users to enter their feedback through the app itself, right? And we
[20:12] wanted to listen to the feedback, right? Then, use the feedback as a requirement, take it back to the engineering team like through automated process, and then allow the process to build the dashboard or make that change to the dashboard,
[20:28] and then deploy it, right? All that we we were able to do that in the, you know, in the 90-day window. Right? Um, that's what we did, and we would like to, you know, get to the demo. I know we've I we've spent too much time in talking,
[20:45] right? Um, let's Satya do the demo. All right. Hi. Can you all hear me clearly now again? All right, just making sure. I'm going to be like going back and forth, with my screen, so please bear with me. Um, just quickly, um, we'll talk about
[21:00] what we'll come cover in today's demo, right? Um, you all may have seen, um, the Databricks experience, you know, in their UI and you know, you know, like going through Genie and whatnot, right? Once we, um, showed that to our senior leadership, that creates kind of a, um,
[21:18] like a perception problem because you're exposing all your data engineering kind of tool to somebody. And the feedbacks, first round of feedbacks was, "Hey, can you package this as like a one-stop shop where, you know, like I don't want to go, um, looking for links. I don't want
[21:34] to go, um, uh, look at the left pane which had all your, you know, data catalog, um, your notebooks and stuff in your UI. Like can you build one experience?" Is what the feedback that we got, right? So, we going to I'm going to walk you through, um, that today. And then number two in the demo what I'm
[21:50] going to cover is, um, the engineering side of the development side of things, right? Like so somebody is like looking for an enhancements or like adding a new feature to an existing dashboard. How is that, um, feedback being passed to the engineering teams? In traditional way, like people hit up in Slack or they set
[22:07] up meetings, um, and that's what we want to avoid and then get into like kind of automation, right? So, that's what we going to cover today, right? Like number number three in the demo what we'll be walking through is is the, uh, is the the version four which Balaji walked through is the true self-service
[22:24] experience, right? Um, where the engine, uh, the consumers are consuming the KPIs from our platform and they look for something which is currently not there. So, they they provide that feedback. Now, how do we take this feedback and then, you know,
[22:39] like pass it to, um, corresponding team which manages that domain data and then have that domain data, uh, team, um, make the development in a much faster way, right? With AI data coming in and daily advancements coming in. For
[22:55] example, we started with Genie code that was able to do magic with our timelines development, right? How do we take even that, you know, um, uh, and add much more tools around our organization, like for example, we have a bunch of MCs MCPs
[23:10] that we want to connect to to to gather context about our own ecosystem, right? For one simple problem is, you know, the the templating, right? Every organization will have a BI dashboard template. How do you take that and then make it a cookie-cutter
[23:26] across all the dashboards, right? The same standards. So, so we were able to take all this and then, you know, cook it into the spec-driven development using open spec and open code and then make AI codify it, right? How can you take,
[23:43] you know, the traditional way of dashboarding, right? Where you have a desktop tool where you drag and drop stuff and then do things. And where you go from there to have, you know, like a Visual Studio Code kind of experience, right? As an engineer, you you tend to write code and configuration and check
[23:59] in to get. And that's what we want to show. And then this is automatically deployed in Databricks using our asset bundle pipeline. So, that's what I'm going to go and then show today, right? All right, this is your traditional Genie experience, right? Where this is
[24:15] the first version that we we came out with, which kind of had all these, you know, like plugins which came up with came along with. And that's where we went into our version one where we tried Databricks one experience, which was a tremendous
[24:30] improvement, right? Now, how do you take even a one step further is like how we try to brand it and package it as our own application, right? So, here is the the commerce insights application which is consumed by our tech leaders within our organization, right? As somebody
[24:46] logs in, it shows them all the dashboards which they have access to and then they have recently accessed based on their their own like access behavior, right? And this gives them one place to go and navigate all their dashboards and
[25:02] gives them like when this dashboard was refreshed and whatnot and who is the owner for this dashboard, right? This gives them uh in in traditional way you have like all links bookmarked and when people like always share links and that always go get lost in transition, right?
[25:19] So now here you have one-stop shop for all your dashboards which um for example is in this is me I have access to, right? And now the second thing which we wanted to show is a marketplace. Here marketplace is is the all the dashboards
[25:35] that our extended team has built together, right? This is all the golden KPIs that our teams have uh compiled and it's available for all the leadership to go and access. And here you can see and come and auto discover um any dashboards that somebody
[25:51] would be interested in and then come back and then request for access. Like for example, let's say I currently do not have do not have access to the sales dashboard. I come and submit an access request. It goes to the corresponding domain team who owns this dashboard and
[26:06] then they make sure that, you know, you are authorized to access this dashboard and then they approve your request. Once you once you're approved, it just shows up in your home page and then you you can access it, right? Going back to the homepage, just wanted to quickly show
[26:22] you the experience of one of the dashboards. So we put together this demo dashboard which talks about all our geo defects, right? And in here is that experience they can navigate into. And one thing that we kind of standardized by talking and taking all the feedback is this kind
[26:39] of a dashboard template where it first it gives me gives a read me kind of a navigating tool to talk about what this dashboard is and a guide to explain what these KPIs are and whatnot, right? Number two is um the standard um
[26:56] bare minimum or golden KPIs uh which we think are going to be crucial uh to tell a data story. And that is all this is about. It's going to talk about KPIs on the top followed by some trends in the bottom, right? So, here in the in this case um it talks
[27:13] about Jira. So, it it it talks about Jira defects. So, here we have want to show the defects and then their status and their resolution rate by team and whatnot. So, we broke it down by as much as granularity to a minimal extent so that you don't uh you don't create that
[27:30] cognitive overload. Like when a user now comes in and say like they want to like ask or drill down more data, instead of they coming to us, they're just going to like go and navigate into this Genie um screen which is going to give them that self-service experience. And they can
[27:46] interact and chat um instead of you know like traditionally going and then asking um somebody what what it would need, right? So, just I'll I'll go through a couple of questions just I I went through here um just just start starting
[28:01] off the demo, right? For example, let's say like when somebody wants to ask the question, like show me um you know like the top five teams by their open different defect volume this this month. Let's say I'm an engineering manager. I do not want this to be on another chart, right? Like you you conversationally ask
[28:18] this question and then Genie comes back with the uh you know um the chart for you, right? One interesting theme uh thing that we were able to solve is earlier we were not able to match the theme which we came up with in the dashboard uh uh against the theme uh for
[28:34] the charts which Databricks uh Genie was coming up with. And that was also recently solved on you know we able to match and then create a consistent experience between what you see uh for the charts and coloring color theme on the left pane versus the right pane. So, that was also like programmatically
[28:49] solved, right? Now, this talks about this this question, it talks about your top five team. Along with the chart, it also gives you an analysis, right? Now, if somebody wants to, you know, like take this and then convert, like for example, I don't want this as a bar
[29:05] chart, I just want this as a line chart, they can just go and ask that question, right? I want to quickly go and then show you like a couple of more examples, right? Show me all the open count for all the release blockers. I this is something our release management leadership would usually ask, and then
[29:22] we we have multiple places earlier to go to, but now they can conversationally ask here and then we can break it down, and then get you the count, right? Now, I want to ask drill down more questions, right? Like, what is my average resolution time for closing these defects
[29:37] for critical and blocker defects versus my medium defects, right? And Genie does it does it, right? It It takes It takes the data and then it compiles and then gets you an output saying like, "Hey, my critical blockers are getting resolved in 14 days." Which is like this
[29:54] is a marked data, by the way, which is pretty bad, but it's able to come back, and and leader can make an informed decision based based on it, right? Like Like I cannot be taking, let's say in real world, I cannot be taking 14 days, so I got to be taking action on on top of this, right? Now, how do you like for
[30:10] example, this other thing is like Now, I have this chart, like can you convert this into a pie chart? That's That's my next question, and it it does that. Like instead of me building those charts traditionally, it would take at least a 3 to 4 days cycle,
[30:25] right? When somebody gives the requirements over a Slack chat or like over an over a meeting or an email, you take that and then you put it against the list of backlog that you already have, and when you come to it, and then when you implement it, it's already like 3 days, right? But now, it gives that
[30:42] like seamless experience for the consumer like where they can ask in moments and get that answer by themselves, right? Now it can do like even even further you know like research oriented questions, right? Now I want to take hey show me my top five teams where
[30:59] the open defect count including their average resolution days and closure rate and then count of release blocker tickets, right? And Genie does that like it does a pretty detailed analysis of what are the key findings and what is my closure rate, what is my
[31:15] resolution performance and what is my release blocker along with an analysis it produces me the it produces me the KPIs. And also it gives the thinking it also shows you what it thought through and then what this experience Let's say like I do not like
[31:32] this and I know for a fact this number is wrong or I do not like the way it presented they're always you know like okay to submit a feedback from here too. For example, let's say something did not work or something went wrong they can submit that
[31:47] feedback over here, right? I do not think this is right. And what we have done as an engineer organization is we tap into all these messages and then make sure that our numbers that we have been providing
[32:03] is good. And the second thing that we also do is we look at all these chat conversations against the sequel that we generated and we in our first cycle we we wanted to you know like make sure that the sequel to the text or the text to sequel generation was accurate. And
[32:20] first time we did it manually and second time what we did was Genie had this benchmarking capability where we were able to put all those like common questions and train Genie in a way that it's able to answer right in a right way. All
[32:36] Now, Now, I want to like go and then show you one more experience, right? Now, let's say like there is there is resolution resolution code RCA is currently not in this dashboard. I say I we did not
[32:52] bring that at all in our data set. And they try to converge here and that information is not there. So, what we wanted to show is how do you communicate that feedback and how do you implement that development, right? So, that's where we implemented this feedback button.
[33:08] Um where I wanted to show you, hey, I do not have uh the RCA here, right? Um this dashboard or Genie does not have RCA field.
[33:26] Add the RCA field and build me a chart. Right? So, you submit that feedback from here. And then that goes to whoever manages this dashboard within our organization, right? And that's you can track your
[33:42] status for your feedback right here. And then what it does is it creates a ticket in our back-end team back-end Jira and it gives you all the details about like what is my dashboard and what is the feedback that was provided. And the development teams take
[33:58] a look at this and then and they want to go implement it. Or let's say this already there, they go and provide that feedback which again goes back platform. So, in this case I do not have it. I have we've not brought this in, right? So, how do you take this and then do that development? And that's where I
[34:14] want to go and flip into Visual Studio Code and then show you how this would be done, right? So, right here I'm showing you my Visual Studio Code here. And then
[34:48] Right? So, what we have done is we have taken Data Bricks CLI along with the agent skills which is also from Data Bricks open source and we are able to you know, like automate this problem you know, this this problem programmatically. Now, what
[35:03] this does is it takes it goes to our Jira MCB, gets all the information about the ticket and then it runs through the skills and the agents are able to go do all three things, right? Which is number one, do an impact analysis and then find out
[35:19] what changes is needed and it proposes a a spec. Number two, what it does is it goes and then fixes all the underlying nuances, right? Number one, which is the metrics view, which is the backbone for all our KPIs. And number two, the dashboard. Number
[35:35] three is Genie space. So, we it does all the nine yards and then implements that functionality, right? For example, just just before this demo, that's what we did. I want to quickly show you in a visual way what what it did, right?
[35:50] Let me quickly pull this. All right. I'm sorry. It's not able to connect to my back end DB. Instead, I can show you in the
[36:07] change, right? For example, this is the spec which it did. And while I'm talking through it, it's still again planning for a new fix.
[36:23] Right? So, here it talks about what all things that it needs to do, um um and then creates a proposal and then um list of tasks that it needs to do. And then based on the Databricks CLI, it's is going to go
[36:38] implement that change um in the in the application. And once what it does is it creates a draft for me and then I can go and then approve. Let me quickly show you and go back to the um application, right? Uh just for the sake of demo, I created
[36:53] one more version. Um if you'd seen my first version, I did not have a chart. Um sorry, let me go back to home screen. Right here.
[37:14] So, I did not had a chart for um RCA. And now based on this feedback, it was able to implement that and then give me a draft which was I was able to manually approve it and then we were able to publish it all end-to-end in a matter of few minutes, right? And that's the scale that we are looking at in terms of
[37:29] bringing down our development times. Right? So, that is number two thing that we wanted to show. And number three is um we built one more thing called knowledge assistant where functional context for example, um
[37:46] the you may have business-related questions. And currently this Genie capability is able to answer most of the questions. But what about like more information? So, this this knowledge assistant is able to answer all those questions.
[38:02] Right? Now, that's all with the demo. I want to quickly um go back to my presentation and then talk through um like Yeah. Talk through the outcomes, right?
[38:26] All right. One of the one of the top takeaways is right like how can we fast track the development timelines? And that that is number one. Number two is how do we you know like augment our engineering team to have this BI narrative skill.
[38:41] And that is a biggest takeaway. Number three is how do you reduce your tool sprawl right? Like for example, as we started we currently have four different BI tools within our organization. And we are already in the works to like kind of decom one or two, which is a big cost
[38:57] savings not just for the license, but also for the operational cost as well, right? That's that's number one. Number two is our development cost right? To to develop something earlier we used to take like few weeks. Now it is a matter of few days and that's that's another
[39:12] biggest takeaway that we have, right? What we what what we have accomplished is like everything in one place, right? Number one, users are good with natural language interactions. And number three is we have all controlled in a secure way using Unity Catalog. Like we have
[39:28] golden minimum dashboards. That's what we've seen in the marketplace where people can go and consume there from. And we have all our KPIs certified against test cases, right? And number three we have all the curated data in Unity Catalog and that's that's our
[39:44] outcome which we were able to accomplish. And then I'll pass it to Balaji and Varghese to give like some closing notes.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.