Sponsored by: Hex | The Evolution of Data Teams with Hex and Rivian

Summary

Rivian faced a fragmented analytics ecosystem of more than 10 tools spanning manufacturing and enterprise segments and ran a vendor evaluation to find a single platform capable of serving both engineers and non-technical business users.
Rivian selected Hex as its unified analytics platform on top of Databricks, enabling conversational analytics, consistent KPI definitions, and self-service access to billions of rows of manufacturing and operational data.
After consolidating onto Hex and Databricks, Rivian achieved consistent KPIs across teams, reduced dashboard sprawl, and built cloud spend monitoring—demonstrating how unified context accelerates organization-wide analytics adoption.

Chapters

FAQs

Why did Rivian need to consolidate its analytics tools?

Rivian had accumulated more than 10 fragmented analytics tools across its manufacturing and enterprise segments, creating inconsistent KPI definitions and barriers to self-service analytics. The team ran a structured vendor evaluation focused on serving both technical data engineers and non-technical business users from a single platform.

What is Hex and how does it work with Databricks?

Hex is an analytical platform that sits on top of data infrastructure like Databricks, allowing users to query, explore, and collaborate on data through notebooks and apps. According to this video, Hex enables conversational analytics and AI-assisted exploration of datasets, including billion-row datasets from Rivian's operations.

How is AI democratizing data access for data teams?

This video explains that AI is blurring the lines between data scientists, analysts, and business users by lowering the technical barrier to querying data and building predictive models. As a result, data teams are evolving into internal AI platform teams whose primary value is providing the context and governance that makes AI-generated answers reliable.

What results did Rivian achieve after adopting Hex on Databricks?

Rivian achieved consistent KPI definitions across teams, reduced dashboard sprawl, and enabled conversational analytics for users who had never written a query. The team also built cloud spend monitoring on Hex and Databricks, turning a previously fragmented analytics environment into a governed, scalable foundation.

Full transcript

[00:08] Hello. Good afternoon, everyone. It's the last day of the summit. Or is Is it Is it tomorrow, too? Yeah, okay. Well, it's almost the last day. It's the second last day. I know it's last day for me. I have to I have
[00:20] to leave tonight. But, uh how's everyone doing? Wednesday afternoon slot can be like a little rough. Is everyone like Did everyone have lunch? You're like, all right. All right, cool. Uh well, thank you for uh coming. For
[00:33] those of you who meant to be here, I'm so sorry and apologies to those of you who wandered into the wrong room. Uh My name is Barry. Uh I'm the co-founder and CEO of Hex. And um I'm really excited to be here
[00:46] today uh with our friends from Rivian. Um This is a slide that Databricks uh told us to put in the deck. Uh Does everyone want to hear me read this or are you guys just good skimming it yourselves? You want me to read it? Okay. This
[01:00] presentation has been prepared for informational purposes only. No, okay. Okay. Well, I'll I'll just skip it. You guys can catch up on it later. That's fine. Okay, I've also been told to tell you to
[01:12] complete your surveys or not. I actually don't care. Um Yeah, I care about the forward-looking statements. I don't care about the surveys. Um and we're here to talk about the evolution of data teams with Agentic
[01:24] Analytics um with our friends at Rivian. So, uh as I said, I'm Barry. That's me over on the right as you've probably figured out. And uh I'm joined here by Justina and uh Saravanan who will introduce themselves in a moment.
[01:37] And um I'm going to set the stage a little bit. So, I'm going to start with something that is um embarrassingly, stupidly uh obvious if you're attending this conference especially. Uh more
[01:49] people are able to ask data questions now because of AI. Big insight, right? Very exciting. Uh and it used to be that you had to have all these technical skills to create a query or a predictive model.
[02:02] Um and with that the line of like who was a data scientist and analyst and just curious people was all very sort of clear. Uh very delineated by technical skill. And of course all of that's blurring now and it's kind of been thrown up in the
[02:14] air. Um these silos and these sort of barriers between these workflows that used to exist have gotten very very blurry. And the barrier for entry to all of
[02:26] these things has gotten very blurry. And I think a good way to describe what's going on in data and AI right now is it's all just sort of very blurry. It's very exciting. Uh but there's actually really that's my
[02:38] reaction but there's actually really two reactions people have. One is this is super exciting. This is great. Everyone can be informed by data. Yay. We hear about that from a lot of our customers. Unfortunately, we also hear another thing from a lot of our
[02:50] customers which is oh no, this is super scary. Everyone can be misinformed by data which is of course the big fear and the risk. Just this morning I was fielding uh talking to a customer at our um space
[03:02] who was talking about how excited they were about all the new features we were turning on but then was in the same breath talking about how they want to be able to turn them off for certain people who they don't trust to ask data questions. This is the tension everyone is feeling today.
[03:16] And regardless of where you fall on the spectrum of excitement to terror, uh whether you're trying to restrict people or whether you're sort of letting it rip, um there are some sort of simple conclusions I think everyone's basically
[03:28] arriving at. AI adoption hinges on accuracy. If someone gets a bad answer, they're probably not going to stick. And I know many of you are probably in roles where you're trying to roll out AI and data tools internally. And probably a big
[03:40] part of your psyche is like, "Hey, are people going to have a good experience with this?" And fundamentally with data a good experience will often come down to their perception of accuracy. Accuracy relies on context. Uh I don't need to tell you that. You've probably heard 57 different talks at this
[03:53] conference talking about context. And I think where the divergence is a little bit is around the workflows on this because I think what we're seeing a lot of is people being like the context that
[04:05] semantic models and the other sources, but um, we think there's these new context workflows that are required and that context should be shared. We think that the context should be universal across everyone using the different data tooling and this should fundamentally be
[04:18] a collaborative workflow. And so this all brings us back to a point and an observation we're seeing amongst data teams. Data teams are being becoming context curators. I actually think about this even more largely for a
[04:30] moment. I think about data teams over time are actually becoming these internal AI platform teams. And I think a huge part of that is context curation. I think the role is shifting from just answering questions to building out the context for other people to ask
[04:43] questions. Um, now this was kind of always the dream but never really achieved. I mean for like a decade at least, more than that, data teams have basically been these service desks. Um, if you're like me when I was when I
[04:55] was in data roles, um, there's a lot of like getting requests from people, building dashboards, sending them back to them. They give feedback on the dashboard. You go and iterate on it for 5 days. You send them back a new version and it's
[05:07] like this this sort of very like get in line version of this. And that's sort of all changing and that's again exciting and scary, but folks aren't bottlenecked on data access anymore. You aren't the only ones anymore with access to query
[05:20] the data or know how to navigate the BI tool. The new bottleneck is going to be on trust. When someone gets an answer they don't believe, forward progress stops, trust breaks down, people are going to find other ways to achieve what they want whether it's just going with
[05:33] gut or hacking together some of their own analysis somewhere else. And so while AI emphatically democratizes the ability to ask questions of data, it also concentrates value in people who understand the business deeply enough to
[05:45] know the questions worth asking and the context to bring together on being able to answer them. Um I think there is a narrative out there that like data teams are going to be less important in the future and it's one we very much disagree with actually.
[05:57] But how these workflows look and evolve is going to change. And we're observing a new thing happening in practice. Now the traditional way to work on this might be familiar to many of you, which
[06:09] is that the data teams basically just go and answer questions and create insights. Those insights become context though now in this new world that people can then self-serve on. They're not one-off lost artifacts, but they are can
[06:21] compound. And then people really can naturally develop their own questions and when they do, they're going back to the data team for help. We think about this as this compounding cycle, a virtuous cycle of data
[06:33] knowledge where the data team is going and creating context, curating context as they work. We think about this is like on a data team often you are tasked with going exploring the new, the novel, and the gnarly.
[06:45] Like what are the things that we don't know yet? What are the new things we're modeling out? What are How is this changing given what's going on in the business? The exhaust from that becomes the context that allow business users to self-serve with confidence.
[06:58] When they hit the wall, when they're hitting a novel question, when they're hitting the limits of the data or context they have access to, that can be volleyed back to the data team and this cycle continues and the whole organization gets smarter and more data
[07:10] informed. Oh, I forgot to hit my uh uh thing to get the the explanatory text up here. Uh but yeah, you basically say, as you're doing your work, what context needs to be updated?
[07:22] Um when agents have all of this all of these things, everyone can actually become a data person. So instead of just looking for the next question to answer, the data team starts looking for where the agents are getting stuck. Where do we have gaps? Where are we missing things?
[07:34] And um we think that this is super super important when you think about how context is not just managed at sort of a raw infrastructure level, but how it's integrated into the actual workflows people are doing. Um we think this compounding cycle is
[07:47] really going to be the key for data teams to unlock in this future. And um it's something we're thinking about a lot. Basically, how do you observe where people are at, what questions people are asking, whether they're getting good
[07:59] answers, and how they can be improved. And uh at Hex, we have the privilege to go work with a lot of amazing teams uh who are doing this all over the world at different scales. And one of the most interesting and forward-looking
[08:12] companies we get the privilege to work with is Rivian. Uh early on in our relationship with Rivian, uh the one of the folks we were working with over there told us, they said their mission was to make data a birthright.
[08:24] And they said unprompted, we want to make everyone a data person. And I smiled cuz that'd been our company motto for a while. And so we knew that we found some kindred spirits. And um I think they've done a wonderful job not only building great products. Um
[08:36] they had a big product launch recently. Um but thinking about how they're running a thoroughly modern and data-informed organization. And so uh I'm very excited to welcome uh two people up to stage that we get a privilege to work with.
[08:49] And I'll actually let you guys introduce yourselves versus me sort of
[09:02] >> My name is Justina Yang and I'm a leader of analytics um and data engineering uh at uh Rivian. And Saravanan? >> Saravanan Lakshmanan. I'll repeat if you're hearing hearing this name for the first time. Saravanan Lakshmanan.
[09:16] I'm responsible for the analytics platforms for Rivian and the sister companies, Rivian Volkswagen Technologies. Right? We take care of all the end last mile needs for the customers. Right? We have multiple tools. We had
[09:29] like so many crazy options. We have boiled it down. We'll walk you through what we have. Okay? With that being said, we have seen our faces. We are right here. >>
[09:41] >> So, this is the opening slide. It's afternoon. You just had your lunch or about to have your lunch before you fall asleep. Let us tell you uh how we got here, how we got into Hex, right?
[09:53] Uh once upon a time, that's how all the story starts. There was this Eric Culture. Is Eric here? Please raise your hand. I'm going to talk smack about you right now. He's a senior director at Rubicon. So, he used his credit card.
[10:06] He didn't even just use the company card. Used his personal credit card and then bought one license of Hex 3 years ago, right? Summer of 2023. None of us knew about it. Shadow IT,
[10:19] right? So, we have the mandate from the leadership, shut every tool down. Okay? We have the systems that we have approved, use those systems. So, we inherited a fragmented ecosystem.
[10:34] You're able to see there are like three different warehouses on the left, right? Extreme left. Uh the the color is so light that you're not able to see, but there are three different data warehouse systems, right? So, we used Snowflake, we used Redshift,
[10:48] then Databricks. Has anybody gone through migrations here? Show of hands. One migration to Oh, you've all gone through the pain. We were going through data warehouse migrations. Then there are 10-plus
[11:00] different data tools, right? All the big shot names, Tableau, Power BI, and there are free tools extremely late, Grafana. Every team were having their own needs. So, a fragmented system that is a lot of
[11:14] tax on us to maintain the platform, give them access, provide security and scalability, governance. It was It was complicated. So, the mandate was there is no new tool coming in, just consolidate the tools.
[11:27] So, we are going through it and then we are segregating into three personas, people who are super technical, like data scientists and stuff, then who are SQL analysts, and low-code no-code users, right? We have to find tools for
[11:39] all these people. Can one tool fix all these things? We figured out like, "No, maybe we need more than one tool." Right? It could be a combination of tools for low-code no-code users, and then one Python SQL
[11:51] heavy tool for somebody else that they can go ahead and then just go, build so many things that they want in code, and then use CI/CD, and then version control, all those things. >>
[12:03] >> The before version was absolutely fragmented. The after version was we have one platform, one Agentic platform, that we are doing deep dive analytics and conversational analytics. Three years ago when Eric Ulcer brought
[12:15] this in into Rivian ecosystem, our idea was like, "No, you're using the existing tools. You're not using Hex." Here, I said it, okay? Because it was a new tool at the point.
[12:27] We didn't know. Um so, we saw the use case of it. Then we agreed, "Okay, 10 licenses, not more than that. Only for your team, because we see the value in it, right?" Hex is not uh the same tool that it is
[12:40] as today. There was no Agentic options back then. There is no token maxing or value maxing, whichever way you want to call it. It was completely different, right? But still we saw the value in it. So, like 10 licenses. Then it slowly grew
[12:53] out and to a point it replaced the other tools that we had. It was not easy, but it did replace. We'll walk you through We'll walk you through the data ecosystem that we have, the journey that we went through, the pain points. If you have any questions, hit us up,
[13:06] okay? So. Why we got rid of the legacy BA tools? Justin, do you want to talk about it? >> Sure. Go for it.
[13:19] Beyond the fact that the team had to support multiple platforms, the cost that came with it as well, became very fragmented ecosystem. So, we had siloed platforms where teams were developing
[13:32] apps and dashboards and within that, their own logic. So, you could have had two people who would ask the same question and got completely two different answers. Um it was very difficult to manage
[13:45] access. Every single tool had a different way of how you provide the entry to to the data. And we wanted to provide one streamlined and very efficient way
[14:00] for everyone. We like to use the birth data birthright access. You just join the company and you should be able to have access to all the data you need to do your job. >>
[14:12] >> Um the limited collaboration, if the logic of specific KPIs lived in the dashboard, it was very difficult for others to go deeper into those analysis. We couldn't
[14:25] share the data in the legacy tools and made it very difficult for us to collaborate between the different domain teams. Um lastly, governance was a big
[14:39] challenge. Imagine trying to fit one set of rules, best practices into multiple BI tools. It was very challenging. Um and I think lastly, just no trust in the
[14:51] data, right? You can have multiple dashboards with the same exact KPI names and they're all different. Data migrations are healthy.
[15:05] >> It's hard, but healthy. >> Yes. Data migrations are very hard. Um we know this, especially for our users. Um if I go to my local Safeway and they rearrange the aisles,
[15:18] I get very frustrated cuz we're creature of habits. we like the way things are, and so changing the way that we do things every day, changing the tools,
[15:30] migrating, that's hard on everyone. Um but the value of migration really isn't a lift and shift. It's not migrating from one place to another. It's to simplify the landscape, create
[15:43] really solid foundation for the data. So, what we focused on, and we can talk about it together, um we focused on migrating and shifting
[15:58] the logic out of the BI tools, out of the applications, out of the dashboards, into data models. So, uh at Rivian we use DBT, and we created all the data models, moved the semantic models out of
[16:12] the BI tools into the data lake. Therefore, it didn't matter which tool we were using, it was all very much contained into um one source of truth.
[16:24] Um we had one set of governance based on that, and it was easier to figure out who owns specific logic, who owns specific business process, and metric definition. Um at the same time, um
[16:38] refactoring takes a while. We found that a lot of um the content was stale. And therefore, we were able to clean up a bunch of stuff, and really focus on
[16:52] the dashboards that are still active and in use, and make it more trustworthy. Okay. >> So, this is Rivian's data landscape.
[17:04] Okay. I see a few taking pictures, so go for it. If you look at it, there are 400 plus data sources, right? It's majorly categorized into three
[17:16] different sections. One would be manufacturing. It's a massive ecosystem. Many small companies club together. Then the second would be enterprise, which is how we run the business, which is not complex, but it's important, right? It's
[17:28] crucial for all the financial decisions. Third one would be fleet, right? Telematics. Uh there was a geospatial session yesterday. If a vehicle Rivian vehicle is going under a bridge, how do you know it's going under the bridge or
[17:40] going to turn left or right on the other road? How do you take the geospatial signal? You have to send the telemetry right away quickly. Is it a low-fi signal, high-fi signal? So, there are massive amount of data coming through from different sources, 400-plus
[17:52] sources, right? Petabyte worth of streaming information. There are use cases for streaming as well as for batch processing. Do you want it right away or can you batch it every night and then send it to the system? There are different use
[18:05] cases. Then, once we get the data, we funnel it to domain-centric catalog and centralized governance, of course, Databricks, right? Now, if you see there are different colors, there's a category color, then we group it down, then we move it to
[18:17] different areas, right? So, product is where the engineering team builds the product, right? Then supply chain is where we get the product the subparts, logistics, that team takes care of that. Manufacturing, of course,
[18:29] the factory. It's a whole microcosm of different uh companies. Commercial is where we sell. And vehicle is where the fleet and enterprise is managing the ecosystem. So, we separate it by domains. Then, we
[18:43] use Python, PySpark, and DBT. Our area is Fivetran, DBT, then Databricks. Okay? Ingestion, transformation into Databricks, then we have Hex at the end for all the visualization purposes.
[18:55] We segregate it by metrics, metadata, and then we give the semantics and context. Our notion is the logic should not live in the application layer. It should be above in the upstream so that
[19:07] everybody will have the same answer irrespective of which agent you're using, which tool you're using, and then calling for a specific table and asking, you should get the same answer. So, then we categorize, and finally we're
[19:19] displaying things in different analytic tools, namely Hex. Questions on this one? No? Good to go? Next slide.
[19:31] >>
[19:44] So, we had to go through different uh rubric analysis of 20 different tools that we compared. 20 different tools. We spoke with 20 different vendors. Okay? It's not easy.
[19:57] More than that, business teams were invested in a specific tool. It's not easy to just rip away a tool and like use this one. It's complicated. They don't like change, right? Just like the rest of the human beings.
[20:09] So, we have to upsell, show the vision, like where we want to go, not the current state, because we have broken the broken ecosystem. 10 different tool 10 plus tools. Look at the load on the administrators. Look at the load on the cost. Every team is
[20:22] going through OPEX, CAPEX, right? So, we consolidating. So, we evaluated multiple tools on the same rubric. We came to an understanding maybe we need more than one tool
[20:35] for low code no code users. This is 3 years back, right? We didn't have context engine context studio, we didn't have the MCP integrations, we didn't have notebook agent at that point. We're just playing. You can have polyglot environment. You can use Python or SQL
[20:48] in your code. And it was fast, right? So, we came to a conclusion of two tools that we can use. We wanted to align architecturally going down the line for future. Will it help us or not?
[21:00] The other tool that we had in the POC, they had the idea of the semantic model should be living within the tool. We did not align with that. So, we chose Hex.
[21:12] They aligned with this architecturally by the future vision, too, and they were innovating quickly for us. So, we made a huge bet of moving into Hex from different tools. It was not an easy move. A lot of resistance from different
[21:25] teams. If you're going pitching an idea, you when you go through a migration, you know what are the things happening. Now, we are migrating warehouses parallelly. We are moving from C uh where Snowflake and Redshift into Databricks. Simultaneously, we are
[21:37] cutting down the number of BI tools that we owned in the system. It was complex. That's the time Hex came into picture, right? >>
[21:49] purpose, as we said, what can Hex deliver as a partner? Will they work with us? Will they share the same vision what we have right now? We found much of a parity there, not uh working against each other. There were growing pains. I'm not saying all hail,
[22:02] and people are sitting in the front row, including the CEO. I'm saying that there were growing pains, right? We overcame We overcame through all those things, and right now we are here sharing our story. >>
[22:15] make everybody a data person, so that they can create insight and own the insight. That's how the tool consolidation started. Still, our team's goal is like everybody should be a data person, and Hex is actually helping us reach that
[22:28] stage, right? Yeah. Moving on. This is the secret to adoption. Christina, do you want to share? >> Yeah. We had a sort of a three-phase approach on how to um influence the adoption. So,
[22:44] the first um phase was to build excitement. We wanted to make sure that um our users and our data consumers understand that it's um the new tool is better.
[22:59] It has a much more proper promising um influence um in the future and they're investing in the right tools that aligns with our strategy as well. We focus on power users. We focus on
[23:11] power users to sort of get them hooked in and play with hacks and sort of spread the word. Later on what we've done is we've used those power users to do train the
[23:23] trainers and um kind of influence every domain that Saravanan mentioned. Um >> It kind of naturally evolved. A few people came forward like hey, we like
[23:35] this tool we would like to adopt and teach others. So a community formed. Naturally it evolved. >> Yes. Um the team uh the power users sort of became the domain experts. So they were our sort of
[23:49] domain spokespeople that helped us drive the migration and the adoption. Um we also partnered very closely with a lot of the eye groups and helped them
[24:02] hands-on migrating a lot of the We have some bad memories from that. >>
[24:14] location. It would be like multiple challenges. Security, governance, access control. You're still using IAM, Entra and we have to figure out like do you want to use OAuth? Do you want to just embed the credentials? Do you want to
[24:26] share with everybody? How do you do auditing? These kind of things came up over and over and over. We have to just follow a methodical pattern and then start the process. >> Yeah. Um At the same time while we were doing this, providing training, providing a
[24:39] lot of we partnered with hacks super closely and hacks actually assisted a lot of our users as well, content creators. But at the same time we were sort of limiting the upgrades, the
[24:51] feature in the legacy tools just to make the uh tools much more promising and that sort of the preferred choice um for everyone. Um we ended up having a
[25:06] migration. We didn't set a company-wide goal, an end date, because we felt like there were domains who were going a lot faster versus some domains had maybe
[25:19] more content or more complex logic and that they would need sort of a different path. Um we did lie and gave um a date that was way earlier than the actual due date that we needed
[25:32] to be done. Um but that gave us a little buffer, 3 months buff- buffer, which by the way we did use. Um but it gave us some buffer to make sure that we are able to complete on time. >> There were stragglers. They want to just
[25:45] use some more time. Then someone give us more time. We want to escalate. We need more time. We need to use this tool even more. So, we we said 6 months now, we're going to end the tool, then we slowly extended. >> Yeah. >> Figured it out. Yeah. >>
[25:57] >> All right. So, Ribbon is about 15,000 employees. About out of that about 5,000 um are on the factory. So, if you take that and you have 10,000 employees that
[26:12] are office-based, 70% of that are current uh users of Hex, which is huge. Um what increased with that is the
[26:24] collaboration. Hex has an amazing way to let you use somebody else's analysis and go deep into the data. So, we've have right now over 1,000 users that are
[26:37] actively creating content, whether it's through building apps or just exploring in the already existing apps.
[26:50] >> So, before, we briefly touched upon it. Like we had dashboard, spreadsheets. That's one major people still love Microsoft Excel, right? Or Google Sheets. That's a data source. People treat it as a data source. Then you have to combine multiple notebooks, SQL cells, spreadsheets.
[27:04] You have to download from one environment, put it on the other one, compare. We have removed all those things right now. See, this is after That's what we have right now. Uh catalog, context and semantics, Databricks apps and Hex. Right?
[27:17] All the logic lives upstream. I'm saying it for the second or third time. Application should be only showing the visualization, a minor changes here and there. But we are doing most of the
[27:30] transformations in the DBT layer, then landing the data in Databricks, and then consuming. This has saved us a lot of time down the line. This has saved us multiple conversation of why your number is different than
[27:42] mine. It's coming from the same data source. What kind of transformations you are doing? My number is different. That has removed all those things. If you go to a specific agent right now and then ask a question like how many cars did we produce today? How many cars did we
[27:54] sell? What is the profit? You just get the same answer. Right? That's the goal. That's what we achieved. >>
[28:06] launched, by the way. If you have not seen it, it's in the expo. You can go, sit, take it feel the car. R1 and R2 standing next to each other. Check it out. So, we can also ask questions. What is the average transaction value for R2 year to date? And then it just gives you
[28:19] the right answer. There is no need to build a dashboard for this and then let it stay there for a long time. It's going to take consume space, then we have to have the admins run the auditing, and then set an aggressive policy of 90 days. Apart from it, we are
[28:32] going to archive and then delete. So, conversational analytics is helping us reduce the amount of dashboards, but still answering the questions that we want. Right? >> Yeah, I would say not every question needs a dashboard.
[28:44] We have a lot of ad hoc one-time questions and in the past somebody would create a tool an app in a legacy tool like Tableau that would live forever.
[28:56] How sometimes you probably experience that as well is no one really has time to go back and clean up stuff that hasn't been used. Or maybe they want to keep it just in case if that leader asked that question again
[29:09] in a year. It makes our platforms messy and not providing the most value. Okay, factory. Yes, perfect. Okay, this is a cool story. Um we've inherited the
[29:23] factory data analytics team about 6 months ago during the R2 ramp. So we were building out the car and trying to figure out how to make the factory shops
[29:36] and line work together. Um when we inherited the team it came with a set of applications. One of them was factory views which was based on the React app. Um where things were just
[29:49] completely hardcoded and it felt like a software development. Our boss Mikey was on the um plane with spotty Wi-Fi. We were trying to answer trying to get this factory views app to answer like okay,
[30:02] how many R2s are we producing? It's not big amounts because we're still ramping up but how many is it? We couldn't do that with that um React app. On the plane within 2 hours with a crazy Wi-Fi,
[30:14] I don't know about the weather and turbulence but he whipped out an amazing dashboard just using Hex adding proper um questions and uh using AI and we have
[30:26] a snapshot of with a dummy data. You want to go to the next one? Um this is just a portion of the dashboard. There's a lot more visualizations against it but like it would take us weeks if not months to
[30:39] refactor that React app to do data same. >> Notebook agents. Feel free to use it. >> Yes. Okay. >> So, our director did hit us with a dashboard like this is what we should be
[30:51] used going forward and look at the numbers that we produced. A React app was changed quickly in a matter of a flight from few hours. >>
[31:03] has evolved our way of working uh consolidating all the BA platforms from 10 plus two to just two at this point. Uh one of the data bricks apps and Hex. It has saved a lot of time and energy. No need to context switch from one
[31:16] environment to the other, download from here and put it over there. We just go to one platform, ask questions, open Slack, ask questions, done. It also increased the data accessibility and collaboration because uh we are
[31:29] controlling the layer using Unity Catalog, which means control is set in one place. Within Hex, we are using internally we're using a connection called restricted restricted plus we have named the different data source connections, but how we share the
[31:41] credentials or we put strict OAuth on behalf of user how we want to use it, that's how we are controlling, right? So, complaints so no BA data lands in somebody else's hand or the finance people should see some other data, HR
[31:53] people should stay with the HR folks. We are using groups and collections actually and then segregating a walled garden for people to go ahead and access the data. EDA, exploratory data analysis, people who have no idea how to code. It's like
[32:06] low code, no code users, right? They can also go ahead and ask questions in plain English. It's like, "Hey, tell me how many tires we have? How many nuts and bolts we need to order more?" Basic questions, right? They can get
[32:18] these answers delivered right to their inbox. What's next for us? Given with the time of 5 more minutes, we understand that we are still not perfect and uh we're seeing that there are more and more agentic workflows
[32:31] scoping in. So, the area that we're focusing right now is like a digital employee, digital identity for an agent. How to control its access, what kind of a scope it needs to have. Can it read or can it
[32:43] write? How do we How do we govern them? How do we restrict them from sharing the exploration to different different Slack channels, for example? If you have access to tables A, B, and C, and the agent is using your token, and you're trying to limit it to just A
[32:56] and B, if somebody's asking a question, you're embedding your credentials, can it access the table C? So, these are the kind of questions we're trying to figure out. It's still not clear. That's what we're working towards. Maybe on future sessions we might have something to show.
[33:08] Right? This is what we're working towards, controlling how the agents function. Barry? One a second. >> Yeah. Okay. Well, okay. Whoa. Uh okay, that was great. Thank you. Um
[33:22] I thought I could ask you guys a couple questions. Um Augustus, I'll start with you, just you know. So, where um where were you operating previously? You guys had a lot of different tools. Like,
[33:34] what were the big What would you say was like the biggest single challenge you guys had when you think about that mission of like make data our birthright? What How would you like sum up like what was the challenge? >> Yeah, I think it's that. It's make data
[33:47] a birthright and make everyone at Rubrik a data person. Um doing that while you have 10 BI tools and multiple data lakes is clearly impossible. So, having this
[33:59] siloed um ecosystem and fragmented ecosystem, lack of governance, lack of quality, that was the challenges that we had to overcome. >> Yeah. And then um
[34:11] one of the use cases I know you guys have done is a little bit of an other one, which is like looking at uh like doing like spend analysis about data warehousing cloud spend. >> Yeah. >> Using the data tool, like how does that
[34:23] work? >> was fun. Is Dana here? >> No. >> Okay. >>
[34:36] point. Um, we are actually capturing information from different streams of how they spend on AWS, how they spend on Databricks, what is the token usage. We We are landing every single data into Hex and have built a consolidated
[34:48] platform of what is the monthly spend, what is the DBU spend, how many licenses are being consumed, are people using what they are being given? So, we are tracking everything using Hex including every other data going inside the
[35:01] enterprise. >> Yeah. >> Very cool. Well, great. Well, thank you both for doing this with us and thanks to all of you for um being here. >> Yep. >> Round of applause. >> Thank you. >> Thank you. >>

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.