Cloudera Evolve 2024 | Abhas Ricky, Cloudera

Clips
News
More from Cloudera Evolve 2024

Abhas Ricky

Chief Strategy Officer

Cloudera

Enterprise AI ecosystem is a key ingredient in Cloudera’s recipe for customer success

Cloudera Inc. has taken a team approach by expanding its enterprise AI ecosystem to include a diverse range of industry-leading AI providers, building a portfolio of enterprise AI solutions for its customers.The data management and analytics firm announced partnerships with Amazon Web Services Inc., Nvidia Corp., and Pinecone Systems Inc. last year. Cloudera extended its joint capabilities further in 2024 with collaborations involving Google LLC., Anthropic PBC and Snowflake Inc.Abhas Ricky, chief strategy officer at Cloudera, talks with theCUBE about the enterprise AI ecosystem.“The idea is AI as a team sport,” said Abhas Ricky (pictured), chief strategy officer of Cloudera

play_circle_outline Maximizing Enterprise Solutions with Ecosystem Partnerships in AI: Cloudera's Collaborations with NVIDIA, Snowflake, Google, Anthropic

play_circle_outline Security and governance considerations in enterprise AI

play_circle_outline Ethical implications of AI, bias, transparency, and data trust

play_circle_outline Future trends in AI technology and potential ecosystem partnerships

Info
Transcript

Abhas Ricky, Cloudera

Abhas Ricky

Chief Strategy Officer Cloudera

The importance of ecosystems in enterprise AI has been highlighted at Cloudera Evolve24. The enterprise AI ecosystem at Cloudera includes partners like AWS Bedrock, NVIDIA, Pinecone, Google, Anthropic, and Snowflake, focusing on AI solutions for customers. Security, model lifecycle management, and operational governance are essential components. Partnerships with Google, Anthropic, and Snowflake enhance hybrid capabilities. Ethical AI is prioritized with a focus on transparency and bias mitigation. Cloudera is looking to add partners to meet evolving customer... Read more

explore Keep Exploring

What is the overview of the enterprise AI ecosystem and what partnerships have been formed to better serve customers? add

What capabilities does the platform have in addressing the growing demand for enterprise AI, especially in terms of managing data assets and ensuring data quality for model training? add

What are some strategies for addressing the ethical implications of AI, particularly with regard to bias and transparency, and how can teams and customers be advised on this important topic? add

What are the three core categories that the speaker believes will be important in the future of AI development and enterprise integration? add

bolt Powered by CUBE AI

Abhas Ricky, Cloudera

search

Rebecca Knight

>> Good afternoon everyone, and welcome back. We are in the final stretch of theCUBE's live coverage of Cloudera Evolve24. I'm your host, Rebecca Knight, alongside with my co-host and analyst, Bob Laliberte. One of the things that has been so clear here at Evolve24 in New York is the importance of ecosystems. We are hearing this messaging from on high on the keynote, and we're hearing it while we talk to guests too.

Bob Laliberte

>> Yeah, absolutely. I mean, in today's ever-increasing disparate and distributed world, it really takes an ecosystem to pull together a comprehensive solution.

Rebecca Knight

>> I see what you did there, Bob. It takes an ecosystem. IT takes an ecosystem.

Bob Laliberte

>> IT takes an ecosystem.

Rebecca Knight

>> With that, I would like to introduce our next guest. He is Abhas Ricky, Chief Strategy Officer at Cloudera. Welcome back to theCUBE Abhas. You are a CUBE alum, an illustrious CUBE alum.

Abhas Ricky

>> I am. I had the opportunity to do this doing data work somewhere in a few years back, so thank you for having me back.

Rebecca Knight

>> Well, we're excited to have you. So you've been at Cloudera for a few years. Before that, you're a veteran of the tech industry as also a consultant. You've had big news this week with NVIDIA and also Snowflake and the AI ecosystem. Can you tell our viewers a little bit about this? We had your colleague Priyanka on earlier to talk about the NVIDIA news. I'd love you to just elaborate on what the news is.

Abhas Ricky

>> Yeah, so just a quick overview of what the enterprise AI ecosystem is. So Cloudera, as you know, serves the world's largest Fortune 2000 companies, and majority of these large companies have specific requirements around scale, security, governance, authorization, etc. So last year we pioneered the enterprise AI ecosystem with AWS Bedrock, NVIDIA, and Pinecone, with the intention being that we can build capabilities jointly to serve our joint customers better. This year we've added more partners, namely Google, Anthropic and Snowflake. And the idea is AI is team sport. There are multiple capabilities that you need. You need a data platform so that you can do the ingestion, you can do the data prep, you can do the data wrangling, and at that point in time, you can serve it up after which you need to do rag applications, fine-tuning, prompting, and a host of capabilities. But you also need to leverage the best-in-class out there in terms of language models. You need to leverage vector databases. So the purpose of the enterprise AI ecosystem is to work together with these crème de la crème companies out there to be able to get to a prescriptive solution provided by the experts on our sites and their sites to make a joint customer successful. And I do think that that is something that is required because as you guys know, there is a model every Monday, there's a framework every Tuesday, there's new benchmark results every Wednesday. So a lot of our customers are asking us that, "Hey, can you be a little bit prescriptive in terms of which model to use for which use case?" So for us at Cloudera in terms of strategy, we're not saying that we will only work with these providers.We do work with a host of other providers and partners as well. Having said then, these are the ones that we have deep integrations with. These are the ones that we have a special relationship with. We go to market jointly as well. And therefore you can be a little bit more prescriptive about the needs of the customer.

Bob Laliberte

>> And that's so important because so many of the enterprises who want to get into this, they're rushing into it, but they may not have the skill sets, they may not understand how things work together. So you coming in with your partners and being able to be prescriptive, being able to give them, this is a tested, validated design, these are the workloads that work in these environments best, et cetera, really gives them an advantage for being able to accelerate that time to ROI, which is so critical for organizations now that are making AI investments.

Abhas Ricky

>> Absolutely, and I have always said this, a lot of organizations wait to get an ROI. It's sometimes too long. AI is one of those technologies where a, it's super capital intensive as you guys know, but also B, innovation is moving so fast that you cannot wait. Sometimes you have to take the leap of faith that is one of those times now. And therefore, not only in terms of skill sets and upscaling your practitioners and teams that you have, you almost have to get down into the project development execution and production rather quickly. And that's where the ecosystem comes into the play because you have validated, tested, authenticated, certified high security models. You also have hardware testing accelerated compute, which allows you to run your scale air workloads on GPUs on a form factor of your choice with the best TCO. But then you also have vector databases for semantics or squaring capabilities. So it's a combination of things that all come together that help you to build your RAG applications or your fine-tuning applications in the future.

Rebecca Knight

>> So you mentioned security. I want to dig into that a little bit because there is a growing demand for enterprise AI, but at the same time, there is a tension to make sure that you are secure, that you are compliant, that you are having good governance, particularly with, I know you work with a range of different companies, but including many that are in highly regulated industries, healthcare, insurance, financial services in general.

Abhas Ricky

>> Telco.

Rebecca Knight

>> Exactly. Telco. So how do you deal with that tension and how do you address the growing demand for enterprise AI?

Abhas Ricky

>> Yeah, there are a series of capabilities that are embedded into the platform. So we have a product called SDX. Think of that as a data plane. You get a 360 view of all your data assets through a single and unified control plane, whether that's for metadata governance, whether that's for lineage, whether that's for provenance and a host of other capabilities around that. But even in so far as enterprise AI is concerned, the biggest challenge is can you have access to high fidelity data that you can trust? Because the most important piece is you have to be able to trust your data that you're going to use to train your model on. And that is one of the core things that we're working with our customers and partners alike. The way we do that is through open data lakehouse, which is powered by Iceberg. There are core security features built into that, but even for our AI workbench. So we have two things which the audience might not be aware of. So we have an AI workbench that you can use in a dev environment, and that has capabilities like applications agents, but that also has multiple studios. So you can have a RAG studio, you can have a fine-tuning studio, you can have a prompting studio that has notebooks that also has accelerated machine learning projects. Think of them as one-click deployments for getting a use case started. So across all of them, there's a core set of operational governance built around that. And we acquired a company called Verta last year, which helps with model lifecycle management and also beefs up our AI operational capabilities around there as well, including governance. But when you get to the production phase, we also have an AI registry and we also add capabilities from a platform like role-based access control, like fine-grained access control, because you want to be able to understand who touched the data, who played with it, what did they change, what did it actually come out to be, what did it actually look like before because you want snapshotting capabilities as well, because you want to know what it looks like once that has been used as part of an enterprise AI application or a cloud application, there might be. So there are a series of capabilities that we're trying to get to with it. And I do think that the stakes are much higher than was before because security is a bigger subject. You guys have all seen manifold instances, example, the New York Times article and the copyright use case that we add with OpenAI and everyone else. So it is an increasingly important subject that is coming more and more as part of the discussions and we're looking forward to solving a lot of them.

Bob Laliberte

>> Got it. Yeah, no, that sounds great. One of the things I wanted to touch upon, when I think about the AI ecosystem that you have created, clearly there's a hybrid component to that. And we heard at the keynote this morning massive growth in cloud, right? Your solution being adopted into the cloud, but also on-prem. And what we're seeing is somewhat of a reflection of that. This year, you added Google, but you also added Anthropic and Snowflake, right? So I wonder if you could touch a little bit on that portion of the hybrid capabilities that your solution has and how that impacts ecosystem partners that you choose to add to it.

Abhas Ricky

>> Absolutely. So there are two parts of the question. So I'll start with the easy one first, which is hybrid. I generally believe we're the only provider for true hybrid solutions. The definition for true hybrid means you have to write once and deploy anywhere in terms of application. So application portability. But if you can move your workloads and applications bi-directionally between private cloud and public cloud without application refactoring cost, that's what hybrid means, and that's what we're focused on. We've been told multiple times that the cost of refactoring application can be 3 to $7 million per app. Any bank, any telco, they have hundreds of apps, so do the math, and therefore that's something that we're enabling through at the very core of what hybrid means to our customers. Now your second question in terms of ecosystem, maybe it'll be helpful for me to give you a little bit of overview of what the customers do as part of their roles that we have, right. So let's start with the founding partners from last year. So the first one was the AWS Bedrock. The idea was you will be able to leverage any model from Cohere or Anthropic leveraging AWS Bedrock service. Using a single click and clatter machine learning, AKA our AI workbench. But we also have NVIDIA whereby we said, Hey, we want to do hardware acceleration because if you want to run scale air workloads on GPUs in a private cloud with the best TCO, you can now do that. And this year with the launch of AI inferencing services using NVIDIA NIMs, you can do this concept of private AI. But I do want to say private AI doesn't mean doing AI in private cloud. Private AI means being able to deploy your data, your compute in public cloud or private cloud or even a desktop. So your data never leaves for a third party vendor in terms of secure and privacy. So then we had Pinecone. Pinecone has lots of semantic search querying capabilities. That was last year. In terms of this year, we've added Google Gemini models from Google Vertex platforms that are more than 150 models of that. There's already a use case M that we have around documents summarization, and we have Anthropic. Anthropic has multiple models. We work with Claude Opus or Claude Sonnet 3.5 and a host of others. We have a series of assistants, the SQL AI assistant, for example, the CML copilot for example. So we're saying we will make the Anthropic Claude model as the default for our customers going through. That doesn't mean we will not work with other models, we will, it'll just be the default going forward.

Bob Laliberte

>> Got it.

Abhas Ricky

>> And then obviously Snowflake. For Snowflake, we announced a partnership through the REST catalog. And then the idea being if I'm an end practitioner leveraging Snowflake today at a customer, I can query any data assets within the Cloudera lakehouse through the REST catalog, which is powered by Iceberg, which will significantly lower the total cost of ownership for our customers. That's one part of it. But then Snowflake have also joined the enterprise AI ecosystem because they have their embedding models, the Snowflake Arctic models that are available through NVIDIA names, and therefore they're available through our AI inferencing service in NVIDIA as well. So across all of these capabilities, the idea being you want to be able to bring the models to the data and not the data to the models. And that is something that we're making it real because you can now deploy applications in a form factor of your choice, but also where the data resides. And that drives security behaviors as well, which majority of the large customers are looking for.

Rebecca Knight

>> As the chief strategy officer, how do you think about the ethical implications of AI, particularly with regard to bias and transparency? How do you advise your team on this? How do you advise customers about this? Because this is a topic of growing importance.

Abhas Ricky

>> Correct. I mean, ethical AI, responsible AI, there are multiple terms around it. As we always have said, a model is being trained on a set of data, and the biggest opportunity is to be able to get a high fidelity data set so that you can get a high fidelity output. But that's also an area of concern because the data asset that you are saying, that's where the trust elements comes through. So in terms of your question, how do we advise our teams? How do we advise our customers? We do say then they should be building modern architectures which allow them to be able to expose the data, for example, for those agentic applications, for those AI applications. I'll give you a real example. We have a customer who are already using agents in production. They're doing 70% or more of the LLM calls through agents. And when we were speaking to them, we said, "Hey, what do you think is the biggest opportunity for us at Cloudera?" And they said, the biggest opportunity is solving our biggest pain point, which is exposing high fidelity data through Cloudera machine learning to our agents because they're building agentic applications for things like tax invoice reconciliation. They're using vision language models for something as simple as a know your customer use case, but they also have multiple agents doing different tasks. And those agents are working together with each other, generating more agents. So over a period of time, the quality of data, but also the bias included into that can have multiplicative effect as you generate more agents, not just regenerate the data as well. So we do want to make sure that you have the processes in place whereby you can have the highest level of authorization capabilities and you have auditing capabilities as well. You'll have fine grain access control, and eventually that will help you with getting to the high fidelity data that we're talking about.

Rebecca Knight

>> Excellent. Excellent.

Bob Laliberte

>> Got it. Got it. That makes a lot of sense. So one of the things I'm always curious when I talk to a strategy officer is obviously looking forward and so curious as you're looking out. Great announcement today. You had started it last year. Going forward, what are some of the things that you are starting to look at to think about who would make a good fit? And I'm not asking you to name names or anything like that.

Rebecca Knight

>> Or if you want to break some news here today.

Bob Laliberte

>> Yeah, if you want to break some news ahead of time, that's fine as well. But when I think about moving forward and you think about the LLMs going to SLMs, the agentic AI, things like that that are going on, how do things in the markets just moving so fast and the innovations are happening so fast, how does that influence yourself as a strategy officer and where you're looking to make those relationships and bring people onto the ecosystem?

Abhas Ricky

>> I say there are three core categories. One is everyone is building multimodal agents, and that pace will only increase. Everyone's doing RAG applications and they'll start doing fine-tuning very soon if they're not doing that already. So there will be people who can provide agentic workflows in an easy-to-use fashion, and there will be a category of people who can provide horizontal agentic building capabilities on top of the platform. There will be a breed of providers for that. And no, unfortunately, I can't share names just yet so I'll stick to the category. So that's one. The second thing is I fundamentally believe that majority of the organizations are moving towards a cognitive enterprise. And a cognitive enterprise is similar to having a neural network, but also having the ability to transcribe multiple complex requirements into simple insights, then to value and then to action. And to be able to do that, one of the core things you need is knowledge bases, because knowledge bases help you with insights. It's not just about the data. You want the knowledge base as well. So there are organizations who are doing knowledge graphs and I think we're looking to integrate with them, and therefore that could be a second category. And obviously the third one is there will be more LNM providers, there will be more infrastructure providers in terms of accelerated compute. Right now, there are a few ahead of the game. Right now, there are a few that are available through services that are not necessarily proprietary and limiting, but over a period of time that will improve. And so I think across these three categories, we will be looking to add partners. If there's anything else, we'll let you know.

Rebecca Knight

>> Well Abhas, we will keep our ears glued to the phone. Thank you so much for this time.

Abhas Ricky

>> Absolutely, was a pleasure. Thank you for having me, again.

Rebecca Knight

>> I'm Rebecca Knight. For Bob Laliberte, stay tuned for more of theCUBE's live coverage of Cloudera 24. You're watching theCUBE, the leader in enterprise tech news and analysis.