Databricks Data + AI Summit 2024 | Rohit Choudhary | Acceldata

theCUBE.net

Clips
News
More from Databricks Data + AI Summit 2024

Rohit Choudhary

Founder & CEO

Acceldata.io

Acceldata highlights data governance and observability amid generative AI shift

Among the many subjects covered at this year’s Data + AI Summit were data observability and the ongoing platform shift tied to generative artificial intelligence. There’s rebuilding and, in some cases, complete redos going on.Enterprises are dealing with cloud transformations, modernizations and AI. But at the fundamental level, there’s still the operability model, according to Rohit Choudhary (pictured), founder and chief executive officer of Acceldata Inc.Acceldata’s Rohit Choudhary talks to theCUBE about the future of data observability.“We’ve got to keep it running, and that has to apply for your data warehouses, for your data lakes and your lakehouses,” Choudhary said

play_circle_outline Increase in data growth

play_circle_outline The Crucial Role of Data Pipelines in the Modern Enterprise: Importance of Governance, Management, and Observability

play_circle_outline The Transformational Impact of Cloud Modernization and the Bifurcation of Budgets for AI and Data Analytics

play_circle_outline The Transformative Role of Data Engineering in Accelerating the Evolution of Generative AI

Info

Rohit Choudhary | Acceldata

Rohit Choudhary

Founder & CEO Acceldata.io

search

>> Good afternoon nerd fam and welcome back to Databricks Data and AI Summit here in sunny San Francisco. My name is Savannah Peterson. Delighted to be joined by co-founder and my fabulous co-host today, John Furrier. John, this has been a really fun two days together.

>> Great day. I mean, it's very technical, it's very nerdy culture here.

>> Super nerdy.

>> Yesterday, we thought it was nerdy keynotes. Even more today with Matillion. It's really good. I mean, Databricks is doing all the right moves, open source, open data formats. Essentially trying to scale up the category and builders, and it's all around governance and it's all about data. And it's a data shift, a platform shift, and it's an exciting time. You don't know which way the wind's going to blow at any given day, but the good news is Generative AI is where it's going. So it's fun to cover-

>> Yeah. Sure exciting-...

>> and fun to talk about.

>> And we're excited to have a CUBE alum, a first timer with us. Rohid, thank you so much for joining us this morning.

>> Thank you so much for having me.

>> How has the show been for you?

>> The show has been excellent. I was just telling you guys before we started rolling is that we are getting about 15% to 20% of the audience come and look at the demos and seriously consider a proposition which allows them to manage their data really well. So it's really exciting.

>> What does that say to you? What does that signal to you when you're getting that significant of a percentage?

>> Well, six years back when we started the company, we thought that there's going to be more data, there's going to be more complexity and more data systems, and there's not going to be enough talent in the market to actually manage those data systems and provide the right business outcomes off the data. And I think that is exactly how the market is playing, with one caveat. We did not expect that data is going to grow so exponentially. We thought it'll grow 2x year over year, but in the last year I've seen data growth of 4x and 5x and now it's suspected that it'll become 7 to 8x in the next two years.

>> What's the value proposition that's driving your business? Is it the governance piece? The observability? Managing the data? What are some of the key features that make you guys successful right now?

>> So I think always the intent for the enterprise has been to manage these different data islands where they have implemented different data systems. The challenge was that governance of the previous generation was very static in nature. It was define it once, catalog all the information and then hand it over to the self-serve analytics folks. But unfortunately what has happened is that that's too far down the supply chain. If you think of data as a supply chain, the places where it gets created is towards the application and the landing zone and the transformation and the ETLs that you do. Pipelines have almost become the lifeblood of the modern enterprise. And so monitoring the data pipelines, making sure that the trains are running on time, data is very reliable and you're doing it at a reasonable cost is something that is now become extremely important, especially in this multi-cloud, multi-technology world.

>> So Savannah and I were talking all yesterday and this morning with the other analysts on theCUBE research is that there's another complex factor that's in the market. I want to get your reaction to this, if you don't mind. Things are changing too.

>> Yeah.

>> So there's rebuilding going on. So there's re-platforming. In some cases complete re-do's and you got to factor in legacy stuff, which now you've got migration tools are getting better. How does that impact your business? Are you guys on that builder side? Are you more on the maintenance side? Where's that action happening? And as people start to rethink, I'm probably thinking you're on the rebuild side given the traffic and to your booth.

>> Absolutely. I think the way to think about it is that as enterprises are dealing with cloud transformations, modernization, all the AI and the cool stuff that everybody's doing, at the fundamental level there's still the operability model. That we've got to keep it running. And that has to apply for your data warehouses, for your data lakes and your lake houses. And I think as long as you can provide that central cockpit view to the data enterprise, I think you're still going to be extremely interesting.

>> So you mentioned that you were predicting that the amount of data would increase 2x, and now we're seeing 4 to 5x. Is that reflected in your customer funnel?

>> Absolutely. Not just in our customer funnel, but the amount of upsell and cross-sell that we are doing off our own products and off our own offerings within our existing customer base. We start with one department, we go to the enterprise within 18 months, and this is unheard of from a sales momentum perspective.

>> Yeah.

>> The kind of funnel growth that we've seen in the last 18 months is just unprecedented. This is sixth year of the business, but in the last 18 months has just taken off.

>> You guys are doing great. I want to ask about Databricks because that's the show we're here. You're on the floor, you're getting some good traction. How is that platform evolving with you and how are you evolving with them? Because they're obviously adding more stuff in. LakeFlow, Unity's open source. A lot of builds, a lot of open source action and trying to make that go faster. What's your connection with Databricks? Explain the relationship.

>> So I have a lot of awareness of the Databricks team because when I was in my previous job at Hortonworks, we used to interact with this team very often, so I know exactly the DNA of this leadership and company.

>> They got Hive going in there too.

>> They have everything. Hive Metastore is now Unity catalog and it's great that they've just open sourced the entire thing. I think the way to think about Spark and Databricks is that it's going to be a crucial component for enterprise compute. It provides great flexibility. It brings out a huge community of data engineers, not just BI analysts and SQL writers, and I think they're doing a great job in integrating these two worlds together. So I think from a fundamental abstraction perspective, this is an important data processing framework.

>> I want to ask you a question. We brought this up with the analyst segment because we are seeing with customers on our research side that there's a bifurcation of budgets right now. Last year they borrowed some other budgets that fund Gen AI This year the budgets are really mainly being approved for Gen AI kind of overall growth. Other budgets that don't have an AI impact are getting some help but not a lot of approval. So you start to see that the question is where's the line between data analytics that falls into the not relevant bucket and where's the line. Because there's a gray area between what's truly going to accelerate Generative AI and what is kind of like maintenance mode. You can look at some data analytics and say, "Well, where is that?" What's your reaction to that? How would you-

>> So I think where the focus is right now in the Generative AI world is completely on the unstructured data, but it is moving very quickly to the structured data side and I think it is happening in small pockets, which is can we actually go and predict some KPIs and look at what these KPIs will be six months out? And there's also the element of the non-Generative AI, which is getting implemented on the structured data side. I think that it'll be a while before people realize that they're ready for it. That's a place where Accelerator plays a big role that you don't get your data house in order, otherwise you won't get garbage out for the garbage that you have, but you're going to get disasters out. I think it'll be about 18 months before we see realistic outcomes of Generative AI. But the investments have to start today.

>> Yeah. You have to start the infrastructure.

>> Yeah.

>> Yeah, absolutely. What are some of the use cases that you're seeing in this flood of customers that are coming through? Are there some trends, consistency verticals, or is it across the market?

>> I think if you were to broadly bucket it, revenue capture, revenue leakage, compliance and finance... financial reconciliation, those are like top three, four use cases. Obviously system reliability, no downtime. Those things continue to be the same case. I think this world is no different than the application development or the APM world. For us, data observability is equivalent to what APMs did for applications. Just give them more reliability.

>> What's interesting is that you mentioned data engineering. We've been speculating that data engineering persona will grow faster with a lot more engineering going on around the foundation, hence the data lakes. As data strategy comes into play, data engineers are going to be actively involved in that. How does that change the career path? Because with AI and now agents coming in, we're going to see AI being infused into the data pipelining and managing a lot of in the weeds details.

>> 100%.

>> You agree with that? Okay. What's the role of data professional now who has to be involved in engineering but also being on the business side? Because as agents come online it's almost like they're an app developer at the same time. So what's your thoughts on this new persona, evolution of the data engineering?

>> It's extremely interesting. So what's going to happen is that the data engineers will continue to be the masters of the data that gets fed into the models, but the application developers or the users of that same data on the consumption side, they will be able to put very quick applications for their own consumption. So for example, if you look at let's say a forecasting model that has to be put together, you already have the agent, you just have to point it to the data set that you have. So I think the speed or the time to insight will reduce dramatically, but it'll be at least 18 months, like I said, even before-

>> So data pipelines are essentially going to be like almost procedure calls, if you will oversimplify it.

>> It will get declarative.

>> It's going to be an operating environment where the pipelines are going to be really critical.

>> Yeah. I mean that's the whole thesis and the premise of Excel data. That data pipelines are the lifeblood of the organization, supplying both structured and unstructured data. That space itself will start evolving, the integrations will become easier and I think a lot more data will go through those pipelines.

>> All right, here's a little bit of a fantasy kind of scenario. Maybe we can play with this riff up a bit.

>> Is this your data fantasy right here?

>> The agents come along, they're going to start to do things.

>> Yes.

>> So Generative AI is a runtime-like environment. It generates things. So do you see a day soon where pipelines will be generated automatically based on some deterministic or some programmable way? In other words, the developers can code Generative AI, but at any given time a new pipeline might have to be built, instantly provisioned, connected, and built.

>> It is absolutely possible, and I'll draw out a scenario for you. Let's say that you do an acquisition of another company and there's a bunch of data that has just come in and you would be able to go and spin up all the consumer data from that database and bring it to the central data platform. It's totally possible provided you have the right governance model, provided you understand the operational metadata and the intricacies of what's the cadence at which you should insert the customer data. You have the right observability on the objects that you're trying to merge.

>> Well double down on that. What are the prerequisites for that scenario?

>> Yeah. I think-

>> Because it's not fantasy sounds like it's going to happen.

>> Yeah.

>> Yeah.

>> What are the prerequisites to make that happen?

>> So first of all, you need to understand the usage pattern that who are the users who are using it? Is it permissible? Is it restrictive? There are a bunch of compliance requirements that go into monitoring or integrating anything before you actually say, okay, look, all of this is flattened out. Second thing I would say is that do you even know the operating environments and are they on the same operability planes? If they're not, then what is the integration environment? Where do we integrate all of that? So there's going to be some element of human in the loop. It's not going to be automated, but I wish-

>> It sounds like data engineering to me.

>> It is data engineering. But just that it'll become a lot more easier.

>> Yeah, yeah, yeah.

>> Yeah.

>> Yeah. A lot faster and let people get back to work.

>> Absolutely. Before the scenario that John mentioned over here, I think a lot of acceleration will come in because of these technologies as opposed to a complete smooth play out.

>> Yes. That's why I thought Jensen Wong from NVIDIA on stage was interesting because his whole accelerated computing paradigm is the underpinnings that's going to fuel a lot more horsepower. So even Ali at Databricks is saying, "Hey, let the best engine win." I mean, he's probably referring to compute engine or any kind of engine. So okay, that accelerated computing is going to come to an acceleration of data.

>> Absolutely.

>> That data layer. Now real time has never not talked about much here because maybe it's a little bit downstream, but real time is important too. What's your view of how real time data is managed in the new landscape?

>> I think things will certainly accelerate because if integration becomes easier, then getting data to the right end points becomes easier for sure. That's like a simple strategy, but I think I'm also equally excited about the quantized models because when the quantized models really start heading home, the specific... The use case specific quantized models will become extremely valuable. So you'll have an accounting model, you'll have an HR model, and those will become equally powerful and you don't need really GPUs for that.

>> So we've been calling that specialty models and I think we actually coined the term small language models in the team-

>> That's right....

>> about a year ago and they brought up on stage. So that was a big part of the keynote this morning, Savannah-

>> Yeah.

>> Was small language models. Because we've been ripping on theCube and our special their domain-specific or data-specific.

>> Absolutely.

>> They're basically saying, "Don't discount those." Of course you would agree with that because you're in the pipeline business.

>> I think we all agree with that.

>> What is that scenario for the small language models? Will they be codified? Will they be wrapped in software? How do those language models learn how to interact with and fuse with maybe other data sets in a generative way?

>> Yeah. I think it'll follow the pattern that we've seen in other industries as well, that there's going to be a generic enough model and then there's going to be specific fine-tuning that the enterprise will have to do with their own data sets and only then it becomes useful. The key things to note over there are can you use the data that you're trying to train it on? Are the outcomes supplying the kind of outcomes that you're generally looking for? What sort of the drift on the data that you were training it on, and is the state of the business the same that it was say three months back?

>> It's an exciting time. I'm curious, since you get to see a lot of different things and talk to a lot of different companies, taking off your cell data hat and just as a human, what excites you most about the next 18 months?

>> I think life is going to get a lot more simpler. We won't have to call people for appointments. I'm just looking for that day when scheduling becomes so much more easier. But I think I'm also equally excited about things like cancer research. Just applying the models on the hundreds and thousands of X-rays that are already available, finding tumors early. Possibly there are acceleration in molecules and chemicals and Medicare. I think there are a host of possibilities that are ahead of us, and I think compute was a big factor in analyzing all the X-rays for tumor detection, for early symptoms of people who suffer from cancer. I think those are some really exciting fields.

>> Yeah.

>> Are we going to do more in space exploration and being able to find more galaxies because we can now process more data and images? I don't know.

>> First we got to get on Mars first. Let's get on Mars.

>> I don't know, we might find there's a cooler place to go even outside of Mars. Who knows? There's a whole bunch of possibilities with high-performance computing getting to where it's at and yeah, I'm excited. We're going to be living longer, happier, healthier lives.

>> I hope so.

>> Yeah. Well, I'm putting that out there. I'm manifesting it. That's at least what I'm trying to do.

>> The first thing I'll give up is end-to-end driving. If AI is reliable, I'm not going to drive anymore.

>> Well, exactly, and it was... GM was a part of the keynote yesterday and with the goal of as a company with zero collisions, obviously it should be a goal for all of us. But to your point, yeah, I'm with you. I mean, Lyft is already as close as we can get.

>> Yeah. I'll test those models in India, by the way.

>> What's next for you guys? What's on your roadmap? Put a plug in for the company. You guys hiring? What's the business like? What's the strategy? What's on your to-do list?

>> Yeah, I think the first thing that we are trying to do is to inform the enterprise that you're going to exist in this changing world for the longest time. Have an operating model, put the right tools and the views in place. Get yourself the right level of visibility so that you can take early action, enable your data engineers to actually spend less time on production issues and send them away to do Generative AI stuff. Give this work to us, obviously, because that's a plug. I think I'm really excited about how this world will play out in a multi-tech, multi-cloud scenario.

>> I put a post on my LinkedIn inspired by John Kreis, a former Hortonworks employee, about what's the lessons learned from the Hadoop days now, in our days. And I just threw it out there and actually got a lot of-

>> You've been getting some mileage out of that baby.

>> Well, everyone came out. Because we've been in this for 14 years. We've been part of that movement. Same question since you were at Hortonworks. What have we learned from Hadoop and where are we seeing that? Is it the simplicity? Is it the operations? Because Hadoop was right on. It's just that we're not crossing over into data analytics. What did we learn from Hadoop that's being solved now that we can ensure confidence?

>> Yeah, there's the third or fourth time I'm referring to. I think the biggest problem with Hadoop was the operations. It was just too difficult to manage. And I'm worried that the modern enterprise has taken on so much that they won't be able to manage these X number of platforms and Y number of use cases. So I think the successful enterprise and back then, even if you look at the Hadoop world, the most successful enterprises figured the operations very early. They applied the best data engineers. They threw millions of dollars into the operability budget and got it working. I think all enterprises will have to start thinking in terms of their operability model. Once you get ahead of that, then this is a jigsaw puzzle. One solution, one platform is not enough. For real time, you'll have another platform. For data science, you'll use another one. For Generative AI studio, you'll do something else. For data preparation, there's going to be another model. For batch processing of data, you'll still need low-cost compute, which probably is on-premise. You have compliance and regulatory needs. It is going to be a multi-Tech, multi-cloud world.

>> It's complex.

>> It is complex. But it's a lot of fun.

>> We're all working to decrease that complexity though. All right, Rohid, last question for you so that we can play it as a highlight before you're on the show next time. Since you're now a favorite return guest.

>> Absolutely.

>> What do you hope to be able to say at the next Databricks and Data AI Senate next year that you can't yet say today?

>> Well, I would like to say that 50% of the Fortune 500 companies are working with us.

>> Love it.

>> I wish I could say that.

>> Well, you're going to manifest that. I'm going to make sure that AI helps us live happier, healthier lives, and we're going to continue our coverage. Rohid, thank you so much for being here. It's been absolutely fantastic.

>> Thank you so much.

>> John, always a pleasure. And thank all of you for tuning in to our fantastic two days of coverage here at Databricks Data and AI Summit in San Francisco, California. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.

Rohit Choudhary | Acceldata

search

>> Good afternoon nerd fam and welcome back to Databricks Data and AI Summit here in sunny San Francisco. My name is Savannah Peterson. Delighted to be joined by co-founder and my fabulous co-host today, John Furrier. John, this has been a really fun two days together.

>> Great day. I mean, it's very technical, it's very nerdy culture here.

>> Super nerdy.

>> Yesterday, we thought it was nerdy keynotes. Even more today with Matillion. It's really good. I mean, Databricks is doing all the right moves, open source, open data formats. Essentially trying to scale up the category and builders, and it's all around governance and it's all about data. And it's a data shift, a platform shift, and it's an exciting time. You don't know which way the wind's going to blow at any given day, but the good news is Generative AI is where it's going. So it's fun to cover-

>> Yeah. Sure exciting-...

>> and fun to talk about.

>> And we're excited to have a CUBE alum, a first timer with us. Rohid, thank you so much for joining us this morning.

>> Thank you so much for having me.

>> How has the show been for you?

>> The show has been excellent. I was just telling you guys before we started rolling is that we are getting about 15% to 20% of the audience come and look at the demos and seriously consider a proposition which allows them to manage their data really well. So it's really exciting.

>> What does that say to you? What does that signal to you when you're getting that significant of a percentage?

>> Well, six years back when we started the company, we thought that there's going to be more data, there's going to be more complexity and more data systems, and there's not going to be enough talent in the market to actually manage those data systems and provide the right business outcomes off the data. And I think that is exactly how the market is playing, with one caveat. We did not expect that data is going to grow so exponentially. We thought it'll grow 2x year over year, but in the last year I've seen data growth of 4x and 5x and now it's suspected that it'll become 7 to 8x in the next two years.

>> What's the value proposition that's driving your business? Is it the governance piece? The observability? Managing the data? What are some of the key features that make you guys successful right now?

>> So I think always the intent for the enterprise has been to manage these different data islands where they have implemented different data systems. The challenge was that governance of the previous generation was very static in nature. It was define it once, catalog all the information and then hand it over to the self-serve analytics folks. But unfortunately what has happened is that that's too far down the supply chain. If you think of data as a supply chain, the places where it gets created is towards the application and the landing zone and the transformation and the ETLs that you do. Pipelines have almost become the lifeblood of the modern enterprise. And so monitoring the data pipelines, making sure that the trains are running on time, data is very reliable and you're doing it at a reasonable cost is something that is now become extremely important, especially in this multi-cloud, multi-technology world.

>> So Savannah and I were talking all yesterday and this morning with the other analysts on theCUBE research is that there's another complex factor that's in the market. I want to get your reaction to this, if you don't mind. Things are changing too.

>> Yeah.

>> So there's rebuilding going on. So there's re-platforming. In some cases complete re-do's and you got to factor in legacy stuff, which now you've got migration tools are getting better. How does that impact your business? Are you guys on that builder side? Are you more on the maintenance side? Where's that action happening? And as people start to rethink, I'm probably thinking you're on the rebuild side given the traffic and to your booth.

>> Absolutely. I think the way to think about it is that as enterprises are dealing with cloud transformations, modernization, all the AI and the cool stuff that everybody's doing, at the fundamental level there's still the operability model. That we've got to keep it running. And that has to apply for your data warehouses, for your data lakes and your lake houses. And I think as long as you can provide that central cockpit view to the data enterprise, I think you're still going to be extremely interesting.

>> So you mentioned that you were predicting that the amount of data would increase 2x, and now we're seeing 4 to 5x. Is that reflected in your customer funnel?

>> Absolutely. Not just in our customer funnel, but the amount of upsell and cross-sell that we are doing off our own products and off our own offerings within our existing customer base. We start with one department, we go to the enterprise within 18 months, and this is unheard of from a sales momentum perspective.

>> Yeah.

>> The kind of funnel growth that we've seen in the last 18 months is just unprecedented. This is sixth year of the business, but in the last 18 months has just taken off.

>> You guys are doing great. I want to ask about Databricks because that's the show we're here. You're on the floor, you're getting some good traction. How is that platform evolving with you and how are you evolving with them? Because they're obviously adding more stuff in. LakeFlow, Unity's open source. A lot of builds, a lot of open source action and trying to make that go faster. What's your connection with Databricks? Explain the relationship.

>> So I have a lot of awareness of the Databricks team because when I was in my previous job at Hortonworks, we used to interact with this team very often, so I know exactly the DNA of this leadership and company.

>> They got Hive going in there too.

>> They have everything. Hive Metastore is now Unity catalog and it's great that they've just open sourced the entire thing. I think the way to think about Spark and Databricks is that it's going to be a crucial component for enterprise compute. It provides great flexibility. It brings out a huge community of data engineers, not just BI analysts and SQL writers, and I think they're doing a great job in integrating these two worlds together. So I think from a fundamental abstraction perspective, this is an important data processing framework.

>> I want to ask you a question. We brought this up with the analyst segment because we are seeing with customers on our research side that there's a bifurcation of budgets right now. Last year they borrowed some other budgets that fund Gen AI This year the budgets are really mainly being approved for Gen AI kind of overall growth. Other budgets that don't have an AI impact are getting some help but not a lot of approval. So you start to see that the question is where's the line between data analytics that falls into the not relevant bucket and where's the line. Because there's a gray area between what's truly going to accelerate Generative AI and what is kind of like maintenance mode. You can look at some data analytics and say, "Well, where is that?" What's your reaction to that? How would you-

>> So I think where the focus is right now in the Generative AI world is completely on the unstructured data, but it is moving very quickly to the structured data side and I think it is happening in small pockets, which is can we actually go and predict some KPIs and look at what these KPIs will be six months out? And there's also the element of the non-Generative AI, which is getting implemented on the structured data side. I think that it'll be a while before people realize that they're ready for it. That's a place where Accelerator plays a big role that you don't get your data house in order, otherwise you won't get garbage out for the garbage that you have, but you're going to get disasters out. I think it'll be about 18 months before we see realistic outcomes of Generative AI. But the investments have to start today.

>> Yeah. You have to start the infrastructure.

>> Yeah.

>> Yeah, absolutely. What are some of the use cases that you're seeing in this flood of customers that are coming through? Are there some trends, consistency verticals, or is it across the market?

>> I think if you were to broadly bucket it, revenue capture, revenue leakage, compliance and finance... financial reconciliation, those are like top three, four use cases. Obviously system reliability, no downtime. Those things continue to be the same case. I think this world is no different than the application development or the APM world. For us, data observability is equivalent to what APMs did for applications. Just give them more reliability.

>> What's interesting is that you mentioned data engineering. We've been speculating that data engineering persona will grow faster with a lot more engineering going on around the foundation, hence the data lakes. As data strategy comes into play, data engineers are going to be actively involved in that. How does that change the career path? Because with AI and now agents coming in, we're going to see AI being infused into the data pipelining and managing a lot of in the weeds details.

>> 100%.

>> You agree with that? Okay. What's the role of data professional now who has to be involved in engineering but also being on the business side? Because as agents come online it's almost like they're an app developer at the same time. So what's your thoughts on this new persona, evolution of the data engineering?

>> It's extremely interesting. So what's going to happen is that the data engineers will continue to be the masters of the data that gets fed into the models, but the application developers or the users of that same data on the consumption side, they will be able to put very quick applications for their own consumption. So for example, if you look at let's say a forecasting model that has to be put together, you already have the agent, you just have to point it to the data set that you have. So I think the speed or the time to insight will reduce dramatically, but it'll be at least 18 months, like I said, even before-

>> So data pipelines are essentially going to be like almost procedure calls, if you will oversimplify it.

>> It will get declarative.

>> It's going to be an operating environment where the pipelines are going to be really critical.

>> Yeah. I mean that's the whole thesis and the premise of Excel data. That data pipelines are the lifeblood of the organization, supplying both structured and unstructured data. That space itself will start evolving, the integrations will become easier and I think a lot more data will go through those pipelines.

>> All right, here's a little bit of a fantasy kind of scenario. Maybe we can play with this riff up a bit.

>> Is this your data fantasy right here?

>> The agents come along, they're going to start to do things.

>> Yes.

>> So Generative AI is a runtime-like environment. It generates things. So do you see a day soon where pipelines will be generated automatically based on some deterministic or some programmable way? In other words, the developers can code Generative AI, but at any given time a new pipeline might have to be built, instantly provisioned, connected, and built.

>> It is absolutely possible, and I'll draw out a scenario for you. Let's say that you do an acquisition of another company and there's a bunch of data that has just come in and you would be able to go and spin up all the consumer data from that database and bring it to the central data platform. It's totally possible provided you have the right governance model, provided you understand the operational metadata and the intricacies of what's the cadence at which you should insert the customer data. You have the right observability on the objects that you're trying to merge.

>> Well double down on that. What are the prerequisites for that scenario?

>> Yeah. I think-

>> Because it's not fantasy sounds like it's going to happen.

>> Yeah.

>> Yeah.

>> What are the prerequisites to make that happen?

>> So first of all, you need to understand the usage pattern that who are the users who are using it? Is it permissible? Is it restrictive? There are a bunch of compliance requirements that go into monitoring or integrating anything before you actually say, okay, look, all of this is flattened out. Second thing I would say is that do you even know the operating environments and are they on the same operability planes? If they're not, then what is the integration environment? Where do we integrate all of that? So there's going to be some element of human in the loop. It's not going to be automated, but I wish-

>> It sounds like data engineering to me.

>> It is data engineering. But just that it'll become a lot more easier.

>> Yeah, yeah, yeah.

>> Yeah.

>> Yeah. A lot faster and let people get back to work.

>> Absolutely. Before the scenario that John mentioned over here, I think a lot of acceleration will come in because of these technologies as opposed to a complete smooth play out.

>> Yes. That's why I thought Jensen Wong from NVIDIA on stage was interesting because his whole accelerated computing paradigm is the underpinnings that's going to fuel a lot more horsepower. So even Ali at Databricks is saying, "Hey, let the best engine win." I mean, he's probably referring to compute engine or any kind of engine. So okay, that accelerated computing is going to come to an acceleration of data.

>> Absolutely.

>> That data layer. Now real time has never not talked about much here because maybe it's a little bit downstream, but real time is important too. What's your view of how real time data is managed in the new landscape?

>> I think things will certainly accelerate because if integration becomes easier, then getting data to the right end points becomes easier for sure. That's like a simple strategy, but I think I'm also equally excited about the quantized models because when the quantized models really start heading home, the specific... The use case specific quantized models will become extremely valuable. So you'll have an accounting model, you'll have an HR model, and those will become equally powerful and you don't need really GPUs for that.

>> So we've been calling that specialty models and I think we actually coined the term small language models in the team-

>> That's right....

>> about a year ago and they brought up on stage. So that was a big part of the keynote this morning, Savannah-

>> Yeah.

>> Was small language models. Because we've been ripping on theCube and our special their domain-specific or data-specific.

>> Absolutely.

>> They're basically saying, "Don't discount those." Of course you would agree with that because you're in the pipeline business.

>> I think we all agree with that.

>> What is that scenario for the small language models? Will they be codified? Will they be wrapped in software? How do those language models learn how to interact with and fuse with maybe other data sets in a generative way?

>> Yeah. I think it'll follow the pattern that we've seen in other industries as well, that there's going to be a generic enough model and then there's going to be specific fine-tuning that the enterprise will have to do with their own data sets and only then it becomes useful. The key things to note over there are can you use the data that you're trying to train it on? Are the outcomes supplying the kind of outcomes that you're generally looking for? What sort of the drift on the data that you were training it on, and is the state of the business the same that it was say three months back?

>> It's an exciting time. I'm curious, since you get to see a lot of different things and talk to a lot of different companies, taking off your cell data hat and just as a human, what excites you most about the next 18 months?

>> I think life is going to get a lot more simpler. We won't have to call people for appointments. I'm just looking for that day when scheduling becomes so much more easier. But I think I'm also equally excited about things like cancer research. Just applying the models on the hundreds and thousands of X-rays that are already available, finding tumors early. Possibly there are acceleration in molecules and chemicals and Medicare. I think there are a host of possibilities that are ahead of us, and I think compute was a big factor in analyzing all the X-rays for tumor detection, for early symptoms of people who suffer from cancer. I think those are some really exciting fields.

>> Yeah.

>> Are we going to do more in space exploration and being able to find more galaxies because we can now process more data and images? I don't know.

>> First we got to get on Mars first. Let's get on Mars.

>> I don't know, we might find there's a cooler place to go even outside of Mars. Who knows? There's a whole bunch of possibilities with high-performance computing getting to where it's at and yeah, I'm excited. We're going to be living longer, happier, healthier lives.

>> I hope so.

>> Yeah. Well, I'm putting that out there. I'm manifesting it. That's at least what I'm trying to do.

>> The first thing I'll give up is end-to-end driving. If AI is reliable, I'm not going to drive anymore.

>> Well, exactly, and it was... GM was a part of the keynote yesterday and with the goal of as a company with zero collisions, obviously it should be a goal for all of us. But to your point, yeah, I'm with you. I mean, Lyft is already as close as we can get.

>> Yeah. I'll test those models in India, by the way.

>> What's next for you guys? What's on your roadmap? Put a plug in for the company. You guys hiring? What's the business like? What's the strategy? What's on your to-do list?

>> Yeah, I think the first thing that we are trying to do is to inform the enterprise that you're going to exist in this changing world for the longest time. Have an operating model, put the right tools and the views in place. Get yourself the right level of visibility so that you can take early action, enable your data engineers to actually spend less time on production issues and send them away to do Generative AI stuff. Give this work to us, obviously, because that's a plug. I think I'm really excited about how this world will play out in a multi-tech, multi-cloud scenario.

>> I put a post on my LinkedIn inspired by John Kreis, a former Hortonworks employee, about what's the lessons learned from the Hadoop days now, in our days. And I just threw it out there and actually got a lot of-

>> You've been getting some mileage out of that baby.

>> Well, everyone came out. Because we've been in this for 14 years. We've been part of that movement. Same question since you were at Hortonworks. What have we learned from Hadoop and where are we seeing that? Is it the simplicity? Is it the operations? Because Hadoop was right on. It's just that we're not crossing over into data analytics. What did we learn from Hadoop that's being solved now that we can ensure confidence?

>> Yeah, there's the third or fourth time I'm referring to. I think the biggest problem with Hadoop was the operations. It was just too difficult to manage. And I'm worried that the modern enterprise has taken on so much that they won't be able to manage these X number of platforms and Y number of use cases. So I think the successful enterprise and back then, even if you look at the Hadoop world, the most successful enterprises figured the operations very early. They applied the best data engineers. They threw millions of dollars into the operability budget and got it working. I think all enterprises will have to start thinking in terms of their operability model. Once you get ahead of that, then this is a jigsaw puzzle. One solution, one platform is not enough. For real time, you'll have another platform. For data science, you'll use another one. For Generative AI studio, you'll do something else. For data preparation, there's going to be another model. For batch processing of data, you'll still need low-cost compute, which probably is on-premise. You have compliance and regulatory needs. It is going to be a multi-Tech, multi-cloud world.

>> It's complex.

>> It is complex. But it's a lot of fun.

>> We're all working to decrease that complexity though. All right, Rohid, last question for you so that we can play it as a highlight before you're on the show next time. Since you're now a favorite return guest.

>> Absolutely.

>> What do you hope to be able to say at the next Databricks and Data AI Senate next year that you can't yet say today?

>> Well, I would like to say that 50% of the Fortune 500 companies are working with us.

>> Love it.

>> I wish I could say that.

>> Well, you're going to manifest that. I'm going to make sure that AI helps us live happier, healthier lives, and we're going to continue our coverage. Rohid, thank you so much for being here. It's been absolutely fantastic.

>> Thank you so much.

>> John, always a pleasure. And thank all of you for tuning in to our fantastic two days of coverage here at Databricks Data and AI Summit in San Francisco, California. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.