In this interview from theCUBE + NYSE Wired: Future of Agents series, Nachiket Deshpande, chief executive officer of Impetus Technologies, joins theCUBE's John Furrier to discuss the launch of the LEAP AI family of products and how context engineering can close the gap preventing enterprise ROI on agentic AI. Deshpande walks through the four components of the LEAP suite — LEAP Logic for data modernization, Context Fabric for converting structured and unstructured enterprise data into actionable context via knowledge graph ontology, an Agent AI Solutions environment and Prism for observability and governance. He grounds the platform in a concrete airline example, showing how layering operational nuances onto AI models drove measurable progress on lost baggage rates that generic models alone couldn't achieve.
The conversation also explores how Impetus is redefining the traditional system integrator model by blending software accountability with services flexibility into a single delivery framework. Deshpande describes three entry points into the platform: the data modernization lens, the use-case-first agent lens and the observability lens for enterprises already managing sprawling and ungoverned agent deployments. He details how LEAP's six-to-eight week cycles allow organizations to absorb new AI capabilities without freezing their architecture — a critical edge given the pace of model and API releases. Central to the discussion is the distinction between context creation and context engineering: context isn't a one-time build but a living discipline that must stay current as enterprise data continuously evolves. From partnering with Databricks, Snowflake and hyperscalers to targeting a future where 50-70% of customer use cases reach production with measurable ROI, Deshpande provides a practical roadmap for enterprises ready to operationalize agentic AI at scale.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: Mixture of Experts Series. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: Mixture of Experts Series.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Register For theCUBE + NYSE Wired: Mixture of Experts Series
Please fill out the information below. You will recieve an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for theCUBE + NYSE Wired: Mixture of Experts Series.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: Mixture of Experts Series. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: Mixture of Experts Series.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: Mixture of Experts Series
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: Mixture of Experts Series. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Deepak Khosla, Impetus Technologies
In this interview from theCUBE + NYSE Wired: Future of Agents series, Nachiket Deshpande, chief executive officer of Impetus Technologies, joins theCUBE's John Furrier to discuss the launch of the LEAP AI family of products and how context engineering can close the gap preventing enterprise ROI on agentic AI. Deshpande walks through the four components of the LEAP suite — LEAP Logic for data modernization, Context Fabric for converting structured and unstructured enterprise data into actionable context via knowledge graph ontology, an Agent AI Solutions environment and Prism for observability and governance. He grounds the platform in a concrete airline example, showing how layering operational nuances onto AI models drove measurable progress on lost baggage rates that generic models alone couldn't achieve.
The conversation also explores how Impetus is redefining the traditional system integrator model by blending software accountability with services flexibility into a single delivery framework. Deshpande describes three entry points into the platform: the data modernization lens, the use-case-first agent lens and the observability lens for enterprises already managing sprawling and ungoverned agent deployments. He details how LEAP's six-to-eight week cycles allow organizations to absorb new AI capabilities without freezing their architecture — a critical edge given the pace of model and API releases. Central to the discussion is the distinction between context creation and context engineering: context isn't a one-time build but a living discipline that must stay current as enterprise data continuously evolves. From partnering with Databricks, Snowflake and hyperscalers to targeting a future where 50-70% of customer use cases reach production with measurable ROI, Deshpande provides a practical roadmap for enterprises ready to operationalize agentic AI at scale.
>> Welcome back. I'm John Furrier, host of theCUBE here at our NYSE studio. Of course, we have a Palo Alto studio connecting Silicon Valley to Wall Street. We're here as part of Impetus Technologies, new product launch and rebrand around Leap AI. We got Deepak Khosla here, chief growth officer, head of business AI. Thanks for coming in. We just had the CEO on talking about the launch, the platform. Very impressive. And this is not a new, new thing. It's an extension of what's been in the works for the company. So you got a lot of trajectory work done with customers on the services side. You understand the workflows. This has been a nice timing. The market wants agents and they got to have the technology to power them.
Deepak Khosla
>> Absolutely, John. Thank you. Thank you for hosting me here and beautiful setting, couldn't be better. But yeah, you are right. I mean, 20 years, 20 years we've been building platforms for our customers. We know the complexity of the data. Till three years back, this platform was used to build analytics use cases, traditional data science, AI use cases. Then came the LLM, ChatGPT, GenAI, and now agentic AI. Our customers are asking, "Is my platform ready to basically run these particular agents? My agents are running, but are they solving the real complicated cases that some of the NVIDIA and the Anthropics of the world are telling me? I have some agents in production, but still I'm not able to scale them. They're not giving me the answers that I want to, or repeatability of the answers. What's wrong?" And in our mind, we were like, "Okay, the models are awesome."
John Furrier
>> You got to get the data, right?
Deepak Khosla
>> That's what, and models will be awesome as you go along. But then the context is required. So Karpathy told this long time back, sometime back. I mean, long in the sense weeks as long in today's .
John Furrier
>> It's like three years, it's one week.
Deepak Khosla
>> Yeah. He said context is the OS, it's the operating system of the LLM. So yeah, we are very excited. We were in this world, we knew what data is and where it sits and all that. And we figured out the gap is not the LLMs. The gap is the context and that's why we want to fill that gap.
John Furrier
>> Yeah. And we've reported on theCUBE and SiliconANGLE and theCUBE Research for the past year and a half that, okay, models create outcomes. But when you have the context and the data done right, it actually delivers execution capabilities, which in turn turns into revenue and ROI, which is now what everyone's talking about. We get the models. And in fact, they've been decoupled from systems so that people can have choice. The emergence of small language models is now called enterprise models. We've also reported that there's been a huge lockdown of proprietary data or company data that has not been... And the models can't crawl what's inside an enterprise. Those are the crown jewels. So now we're starting to see the formation of the Leap AI product that you guys have seems to be targeting that unlock of data but not throwing away using models. Yeah, we'll still use models, but we got our data. Talk about that. Is that right where you guys are targeting that use case? Is that the main driver? There's data unlock. Are there other things that are involved in-
Deepak Khosla
>> Yeah, yeah.
John Furrier
>> Is it agents? What's the key focus?
Deepak Khosla
>> So three things, three things. The first thing is the data estate itself. I mean, the data which is required for these agents to work is quite different from what typically a data estate and organization will have because now we are talking a lot of unstructured data, we are talking about SOPs, the policy, the procedures, the business rules. We are talking about, I don't know, hundreds of years of legacy of an enterprise which needs to come in because these agents needs to kind of represent, have the agency of an enterprise of a company. So all that data needs to come in. So that's the first unlock we'll do. We'll basically understand why you are basically building this data platform, why are you doing modernization, migration? What's the end outcome? You're bringing all your data in. Has to be some agentic solution. What are you building? Is it customer service? Is it supply chain? Is it underwriting? Is it fraud? Based on that, we'll try to see what kind of ontology, what kind of semantic layer is required for those agentic solutions to work. So right up front, John, what you're going to do is we'll make sure the right context gets moved to the data platform. So that's the first unlock we'll do. Now you have the data. Data tells AI that this is what exists. What it does not tell us is how to take intelligence out of it. And that's where the second unlock will come into play. We'll talk about how we would build the knowledge graph, the anthology layer, the memory layer, the signals. So that's the second level of unlock that we would bring in terms of building those contexts that agents will use. And then now agents itself building, we believe there's a new methodology. I think the old methodology of building IT solutions, AI solutions, and agentic solutions are no longer going to work over here. So we have kind of developed a new methodology, which is called CEDL, Context Engineering Delivery Life Cycle, in which you basically create the context is that for C, you engineer the context, that's for E. Then you have this whole learn model, which is basically you would learn what's happening, what's not. And then you kind of bring back the signals and make those agents work again. So that's a third unlock that we've got to do.
John Furrier
>> I asked Nachiket what kind of business was he in and I said I lead the witness, I should have said, "Hey, you're in the context generation." No, no, we're in the context engineering business.
Deepak Khosla
>> That's right.
John Furrier
>> But data context has changed. So you have data context, that's always changed. You have existing data that's in there, but new data's coming in, engagement, all that's feeding into the systems of execution.
Deepak Khosla
>> Exactly.
John Furrier
>> So explain to me what the context engineering premise is about because to me that's a high order bit around governance, but not just governance, observability, semantic layer. You have all the piece parts of the puzzle. So it seems like you guys are kind of like bundling in all those hard pieces that everyone's doing upfront that they have to do upfront. It's built in from the beginning. Everyone's talking governance, but it's not just governance. What is this context engineering paradigm about?
Deepak Khosla
>> Five things. First thing is context engineering will start building knowledge graph. You kind of build the relationship between entities, who knows what, what gets actioned where. So that's a knowledge graph that you would build. Second is the ontology layer itself. The ontology layer is the business tools. How much discount do you give when? What kind of action you take when? So that's the ontology layer that gets built. The third thing that we build is memory. Memory is very important because you need to have short-term memory, long-term memory, sportic memory. Memory is even more important because the way LLM works, if they learn a wrong thing, by mistake you kind of make them learn the wrong thing, that's going to stick with you. That's going to stay with all the actions that's going to happen. So how do you make this-
John Furrier
>> By the way, on the memory piece, I will add, just get your thoughts, context routing has been a big discussion. Is that included in that memory piece?
Deepak Khosla
>> Yes, absolutely. And I'll tell you the second point of it. Memory is getting costly. Memory is not getting cheap. I mean, you know what's happening over here. And that's number one. I mean, that's definitely number one. Second, I think inference has been highly subsidized by all the AI labs. Now they're saying, "All right, give me the money which is being spent over there." So if that subsidy is gone, you heard what the Uber CTO said, three months, 12 month budget already spent. So now you need to have memory, but also serve that kind of a layer-
John Furrier
>> By the way, on the inference, I'm writing a post. I mean, if you come up on LinkedIn this morning, actually inference costs are dropping, but the compute costs are going through the roof.
Deepak Khosla
>> Absolutely.
John Furrier
>> So it's not just token costs that are going like this. Compute costs are going up. Managing that context, routing or prompt routing or whatever you want to call it is a money saver because you don't want to route something, "Hey, what's the weather in New York City?" To Vera Rubin. You want to go down to maybe... So there's all kinds of policy around how to handle the data in context.
Deepak Khosla
>> Exactly, exactly. That's important. I mean, it's what models do you use? When do you go? You took this example of what's the weather in New York, I'll go and spend some tokens and then somebody again in your family will ask, "Do I carry an umbrella?" Same answer, same token, right?
John Furrier
>> Yeah.
Deepak Khosla
>> So imagine this at an enterprise scale and the kind of work which is happening. And remember now the agents are running 24/7. They're no longer kind of... So having low token costs doesn't matter at all. It's about the total spend that's going to come over here. And just to complete your previous question, so I talked about file layers, I talked about memory. The fifth is signals.
John Furrier
>> Signals?
Deepak Khosla
>> Signals. That's where I don't want to overextend ourselves, but yeah, the AI labs are thinking about AGI, that's something which you see and you say, "Okay, this is the action that I need to..." That's a human mind. Those are the signals that we are kind of building on top of what we are kind of the context and everything that's there. Yeah. I mean, we are there, but that's a key part. And then the fifth and the most important element is the governance, is the trust, is the compliance. So these are the five unlocks.
John Furrier
>> This is the road when you're talking about driving a car. You got the technology's the car, the car is getting better, faster, ready. It's the drivers and they're driving like 16-year olds stealing their parents' car and there's crashes is where we're seeing some innovation. Execution is happening. Shadow AI is really, I think a feature, not a bug. I see a lot more of that. Talk about that business side, because you're the chief growth officer. You have to go out and get customers to engage, deploy the platform, put it into practice, get the execution, get the ROI. How are you seeing the ROI equation? Again, models are good for outcomes. Seeing a lot of interactions there, but when you start into the systems game, and when you add context and intelligence, you're executing differently and everything under the covers. How do you explain that to an enterprise? What's the strategy?
Deepak Khosla
>> Right, right. So two things. Number one, from the growth office perspective, I work very closely with partners. I'm going to answer your question in a minute, but it's very important to say I work very closely with our partners. So all the hyperscaler data product companies, we work very closely with them. Whatever we are building is kind of very tightly integrated what these guys are also kind of giving to their customers. And if you see, I did this analysis for one of our board meetings in April, the kind of new product launches that they have done, all the hyperscalers and the product company, if you see in the last six months is around context and memory. So they are also going that particular path and we are very happy. So now it's not about the discussion that we do. It's not about how do we go together. It's about the shared product roadmap. So that's a shared mindset that we have with them. So that's beautiful.
John Furrier
>> A lot of co-designing going on.
Deepak Khosla
>> Exactly. Exactly. Not co-designing with them, but the way we are building-
John Furrier
>> Co-collaboration.
Deepak Khosla
>> Co-collaboration. So now they know there's an impetus which understands this particular problem that our customers are having. They are building both their platform strength as well as services strength and they can basically understand our product well and implement that for the customer. To do what? Your previous question, two things. Either reduce the cost or increase the growth. That's the only two things which happens. Of course, customer loyalty is also very important. So eventually what we want to see is we want to see there's these agentic solutions which are there. Number one, they're not only the chatbots. Chatbots are good, great, very first unlock, needs to be done, but now are we thinking about use cases for manufacturing companies where we see their claims, merge them with the telemetry data, try to shift lift in terms of the production design side itself, this is what is happening. I'm sure Nachiket would have talked about the airline example, how do we figure out the various parameters that happens to make sure John's luggage reaches with John at the same time? How do we make sure that retail companies who want to basically do launches quite faster, the whole CDP process?
John Furrier
>> The airline example because, and I pointed out this to in a lot of conversation here, is that it used to be that you'd have to bend to the technology now that the technology bends to the use case.
Deepak Khosla
>> Right. Exactly.
John Furrier
>> And so there's no general purpose use cases anymore.
Deepak Khosla
>> Exactly.
John Furrier
>> They're all specific or tailor made or fit for purpose.
Deepak Khosla
>> Right.
John Furrier
>> That seems to be... Do you see that too with this?
Deepak Khosla
>> Yeah, absolutely.
John Furrier
>> All right. So what is the strategy for the engagement with the customer? Because a lot of enterprises are architecting a generational play here. So there's an entry strategy, there's problems to solve, but those problems have to be enabling to keep that headroom for growth because it's a bridge to the future that they're building. They want to have a 10-year horizon on the architecture. What's the pitch? I guess what's the value proposition when you come into enterprise say, are you solving the data problem first? Is it more holistic? What's the approach that you take with Leap?
Deepak Khosla
>> So the way we have set up, or I have set up, the whole offer strategy is we'll meet customers where they are. And in my mind at this point of time, they're at three places. Number one, customers are like, "All right, we did some experiments with gen AI and all that seems all right. We want to go big in, but we don't have that data estate. Can you help modernize, build our AI-ready data platform?" So over there, we'll kind of focus in terms of how do we bring AI, ontology, semantic thinking at the time of building the platform? A lot of my customers come for this whole modernization migration space, right?
John Furrier
>> On the data?
Deepak Khosla
>> On the data, building the data estate. Again, it's not about semantic data. It's not about the enterprise data. It's about the unstructured data and that's what is important. So number one, customers want and check with us, "Is my platform AI-ready? Or I want to basically launch this use case we are sold in terms of agentic customer service, agentic underwriting. What do I need to do in terms of having that data set?" So that's number one. We help them bring that context in while we are migrating and modernizing and building their data platform. Second is that my customer says, "I've invested a lot. My data platform is ready. I think I've got the right AI tools. I'm ready to launch good agentic solutions, solving good problems, not chatbots and all that, solving good problems in production at scale."
So that's where we'll come. We'll start building the knowledge layer, the semantic layer upfront. We'll not start thinking about it. Again, we'll basically bring the best of that enterprise knowledge to those agents. So that's the second point of interaction. Third, customers would say, "I've invested. I have almost everything, but my agents are not working. They're in production. They're not giving me the right kind of ROI." That's where, and most of the times, believe it or not, it's the lack of the context which is kind of stopping the agents to do what they need to do. LLMs are smart and they are very, very smart. But it's like you do a new hire, a new hire in the company, a smart person, and you don't tell them anything about the company, the smart person's not going to do anything well. But if you tell the smart hire about your businesses, your rules, and everything, the guy will do good. So that's the third point of incision, we make those agents smarter by bringing the context there in it.
John Furrier
>> Deepak, I love the use cases there and I think that's instructive. In fact, I've heard a variety of viewpoints, but I think it's general consensus in most senior leadership is, "Hey, let's go after the hard problems first because it's needle-moving." And so a lot of people want to go after not the easiest low-hanging fruit. "Hey, what's the low-hanging fruit?" "Oh yeah, we'll do a chatbot over here. Test it." I've seen companies get paralyzed by too many tests. They do too many low-hanging fruit use cases. I've seen a success where they go, "Hey, we know this use case and this workflow and the data set behind it actually creates revenue. Let's go there first." Because that's not like an IT project mentor. That's a CEO, CFO kind of approach. If that works there, then what do you need for budget? Okay. So take us through that because I think your second example pointed that because I think that's not a motion that you've seen in the past, unless it was a monster IT transformation project with massive budgets, big timetables. We've seen all those SAP deployments and the Oracles and all the big systems that have been rolled out in other generations now happening here, but different. Talk about what you guys do there because I think this is a sweet spot.
Deepak Khosla
>> It is. It is. And I think number one thing to do is to find out the biggest blocker that you have, the unlock that one needs to do. Imagine an oil and gas company which kind of moves their assets all along. It takes them about 80 days, multimillion dollar kind of a thing. It's a big, big area which needs to be unlocked. Imagine the whole underwriting process where you're going after customers. It takes you about five days, four days. Can you basically reduce that? Imagine an unlock in terms of a particular process which is very, very heavily human-oriented and can you basically make that little bit more automatic, autonomous, so to speak. So number one is the unlock that we basically go after. Second thing that we do is while that unlock would be complicated, the unlock would be big. We try to break it down. We are big believer of the skill concept. So we want to break down each particular scenario agent into kits first and kits right below the skill level and then start building those skills. Those skills, let's say even for a fraud use case, which is big use case, could be to figure out what is false positives to a particular human-in-the-loop? So just a skill which can kind of figure that particular thing out. So start building those skills and those skills will turn into kits, those kits are already deployable, and those kits then you can engage in an agent ,and then you can think how much level of agency you want to give to that particular agent to be autonomous or not. So that's how you kind of break now the bigger problem and still you're attacking the bigger problem. You're still bringing those CEDL lifecycle methodology that I talked about and you try to go to production.
John Furrier
>> I just wrote a post, Substack, on LinkedIn and SiliconANGLE on the term I call hyperscaler 3.0. And we're kind of in 2.0 now, which is enterprise adoption, neoclouds. 3.0 is more of a frame around what we see happening. And I want to get your reaction to this because you hit on some of this, which is you have a fully distributed hybrid cloud infrastructure with now an angentic layer that's infrastructure with AI-native agents and builders on top building value. So I wrote the blueprint for the next phase of the enterprise is architecture, receipts, which is proof, and outcomes. What's your reaction to that?
Deepak Khosla
>> Oh, absolutely. I think the kind of investment that's going in, I think there would be a come-to-Jesus moment that, "Show me the output. I mean, all the tokens that I'm spending, what's going to come out of it?" So definitely I think that's what I think 3.0 is going to be, which is more in terms of outcome, more in terms of ROI. I believe that ROI would be to start with more in terms of perhaps efficiency, growing markets. If you get the right context, it's about having that poll position in terms of the new world that you are in. There would be a lot of consolidations. You want to be the one who does the consolidation and who shouldn't get consolidated and I think AI is going to help all that.
John Furrier
>> Okay. So how would you describe, the final question for you, Impetus Technologies' blueprint for the enterprise, what's your pitch?
Deepak Khosla
>> I think leap.ai, we come, we help your agents work better. We help build the enterprise context for your agents that you have. We bring a platform approach, we bring our FDE approach, and we work with hyperscalers. We work with your landscape that's there. So time to market.
John Furrier
>> Time to market is your hyperscaler 3.0 solution.
Deepak Khosla
>> Yeah.
John Furrier
>> This is where it's at. Well, congratulations. I love the timing of the launch. It just wasn't invented to fit the fashion of the market. You guys have done the work with the services with the beachhead. You've been in the front lines getting those requirements, building a trajectory. Let's see how it goes and wish you the best of success.
Deepak Khosla
>> Thank you so much, John.
John Furrier
>> Congratulations.
Deepak Khosla
>> Thank you.
John Furrier
>> I'm Jean Furrier here. The launch of Impetus Technologies Leap AI. This is the leaping of the next generation. Again, the architecture's changing. It's evolving. Architecture-proof receipts and the outcomes ultimately is what people are focused on. Apply the AI to where the money's being made, where the ROI is. That's where the focus is. A little bit different than the IT projects. Obviously, the CFOs involve a lot more of the C-suite. We're doing our part here in theCUBE to bring that to you from the NYSE Wired program. Thanks for watching.
>> Welcome back. I'm John Furrier, host of theCUBE here at our NYSE studio. Of course, we have a Palo Alto studio connecting Silicon Valley to Wall Street. We're here as part of Impetus Technologies, new product launch and rebrand around Leap AI. We got Deepak Khosla here, chief growth officer, head of business AI. Thanks for coming in. We just had the CEO on talking about the launch, the platform. Very impressive. And this is not a new, new thing. It's an extension of what's been in the works for the company. So you got a lot of trajectory work done with customers on the services side. You understand the workflows. This has been a nice timing. The market wants agents and they got to have the technology to power them.
Deepak Khosla
>> Absolutely, John. Thank you. Thank you for hosting me here and beautiful setting, couldn't be better. But yeah, you are right. I mean, 20 years, 20 years we've been building platforms for our customers. We know the complexity of the data. Till three years back, this platform was used to build analytics use cases, traditional data science, AI use cases. Then came the LLM, ChatGPT, GenAI, and now agentic AI. Our customers are asking, "Is my platform ready to basically run these particular agents? My agents are running, but are they solving the real complicated cases that some of the NVIDIA and the Anthropics of the world are telling me? I have some agents in production, but still I'm not able to scale them. They're not giving me the answers that I want to, or repeatability of the answers. What's wrong?" And in our mind, we were like, "Okay, the models are awesome."
John Furrier
>> You got to get the data, right?
Deepak Khosla
>> That's what, and models will be awesome as you go along. But then the context is required. So Karpathy told this long time back, sometime back. I mean, long in the sense weeks as long in today's .
John Furrier
>> It's like three years, it's one week.
Deepak Khosla
>> Yeah. He said context is the OS, it's the operating system of the LLM. So yeah, we are very excited. We were in this world, we knew what data is and where it sits and all that. And we figured out the gap is not the LLMs. The gap is the context and that's why we want to fill that gap.
John Furrier
>> Yeah. And we've reported on theCUBE and SiliconANGLE and theCUBE Research for the past year and a half that, okay, models create outcomes. But when you have the context and the data done right, it actually delivers execution capabilities, which in turn turns into revenue and ROI, which is now what everyone's talking about. We get the models. And in fact, they've been decoupled from systems so that people can have choice. The emergence of small language models is now called enterprise models. We've also reported that there's been a huge lockdown of proprietary data or company data that has not been... And the models can't crawl what's inside an enterprise. Those are the crown jewels. So now we're starting to see the formation of the Leap AI product that you guys have seems to be targeting that unlock of data but not throwing away using models. Yeah, we'll still use models, but we got our data. Talk about that. Is that right where you guys are targeting that use case? Is that the main driver? There's data unlock. Are there other things that are involved in-
Deepak Khosla
>> Yeah, yeah.
John Furrier
>> Is it agents? What's the key focus?
Deepak Khosla
>> So three things, three things. The first thing is the data estate itself. I mean, the data which is required for these agents to work is quite different from what typically a data estate and organization will have because now we are talking a lot of unstructured data, we are talking about SOPs, the policy, the procedures, the business rules. We are talking about, I don't know, hundreds of years of legacy of an enterprise which needs to come in because these agents needs to kind of represent, have the agency of an enterprise of a company. So all that data needs to come in. So that's the first unlock we'll do. We'll basically understand why you are basically building this data platform, why are you doing modernization, migration? What's the end outcome? You're bringing all your data in. Has to be some agentic solution. What are you building? Is it customer service? Is it supply chain? Is it underwriting? Is it fraud? Based on that, we'll try to see what kind of ontology, what kind of semantic layer is required for those agentic solutions to work. So right up front, John, what you're going to do is we'll make sure the right context gets moved to the data platform. So that's the first unlock we'll do. Now you have the data. Data tells AI that this is what exists. What it does not tell us is how to take intelligence out of it. And that's where the second unlock will come into play. We'll talk about how we would build the knowledge graph, the anthology layer, the memory layer, the signals. So that's the second level of unlock that we would bring in terms of building those contexts that agents will use. And then now agents itself building, we believe there's a new methodology. I think the old methodology of building IT solutions, AI solutions, and agentic solutions are no longer going to work over here. So we have kind of developed a new methodology, which is called CEDL, Context Engineering Delivery Life Cycle, in which you basically create the context is that for C, you engineer the context, that's for E. Then you have this whole learn model, which is basically you would learn what's happening, what's not. And then you kind of bring back the signals and make those agents work again. So that's a third unlock that we've got to do.
John Furrier
>> I asked Nachiket what kind of business was he in and I said I lead the witness, I should have said, "Hey, you're in the context generation." No, no, we're in the context engineering business.
Deepak Khosla
>> That's right.
John Furrier
>> But data context has changed. So you have data context, that's always changed. You have existing data that's in there, but new data's coming in, engagement, all that's feeding into the systems of execution.
Deepak Khosla
>> Exactly.
John Furrier
>> So explain to me what the context engineering premise is about because to me that's a high order bit around governance, but not just governance, observability, semantic layer. You have all the piece parts of the puzzle. So it seems like you guys are kind of like bundling in all those hard pieces that everyone's doing upfront that they have to do upfront. It's built in from the beginning. Everyone's talking governance, but it's not just governance. What is this context engineering paradigm about?
Deepak Khosla
>> Five things. First thing is context engineering will start building knowledge graph. You kind of build the relationship between entities, who knows what, what gets actioned where. So that's a knowledge graph that you would build. Second is the ontology layer itself. The ontology layer is the business tools. How much discount do you give when? What kind of action you take when? So that's the ontology layer that gets built. The third thing that we build is memory. Memory is very important because you need to have short-term memory, long-term memory, sportic memory. Memory is even more important because the way LLM works, if they learn a wrong thing, by mistake you kind of make them learn the wrong thing, that's going to stick with you. That's going to stay with all the actions that's going to happen. So how do you make this-
John Furrier
>> By the way, on the memory piece, I will add, just get your thoughts, context routing has been a big discussion. Is that included in that memory piece?
Deepak Khosla
>> Yes, absolutely. And I'll tell you the second point of it. Memory is getting costly. Memory is not getting cheap. I mean, you know what's happening over here. And that's number one. I mean, that's definitely number one. Second, I think inference has been highly subsidized by all the AI labs. Now they're saying, "All right, give me the money which is being spent over there." So if that subsidy is gone, you heard what the Uber CTO said, three months, 12 month budget already spent. So now you need to have memory, but also serve that kind of a layer-
John Furrier
>> By the way, on the inference, I'm writing a post. I mean, if you come up on LinkedIn this morning, actually inference costs are dropping, but the compute costs are going through the roof.
Deepak Khosla
>> Absolutely.
John Furrier
>> So it's not just token costs that are going like this. Compute costs are going up. Managing that context, routing or prompt routing or whatever you want to call it is a money saver because you don't want to route something, "Hey, what's the weather in New York City?" To Vera Rubin. You want to go down to maybe... So there's all kinds of policy around how to handle the data in context.
Deepak Khosla
>> Exactly, exactly. That's important. I mean, it's what models do you use? When do you go? You took this example of what's the weather in New York, I'll go and spend some tokens and then somebody again in your family will ask, "Do I carry an umbrella?" Same answer, same token, right?
John Furrier
>> Yeah.
Deepak Khosla
>> So imagine this at an enterprise scale and the kind of work which is happening. And remember now the agents are running 24/7. They're no longer kind of... So having low token costs doesn't matter at all. It's about the total spend that's going to come over here. And just to complete your previous question, so I talked about file layers, I talked about memory. The fifth is signals.
John Furrier
>> Signals?
Deepak Khosla
>> Signals. That's where I don't want to overextend ourselves, but yeah, the AI labs are thinking about AGI, that's something which you see and you say, "Okay, this is the action that I need to..." That's a human mind. Those are the signals that we are kind of building on top of what we are kind of the context and everything that's there. Yeah. I mean, we are there, but that's a key part. And then the fifth and the most important element is the governance, is the trust, is the compliance. So these are the five unlocks.
John Furrier
>> This is the road when you're talking about driving a car. You got the technology's the car, the car is getting better, faster, ready. It's the drivers and they're driving like 16-year olds stealing their parents' car and there's crashes is where we're seeing some innovation. Execution is happening. Shadow AI is really, I think a feature, not a bug. I see a lot more of that. Talk about that business side, because you're the chief growth officer. You have to go out and get customers to engage, deploy the platform, put it into practice, get the execution, get the ROI. How are you seeing the ROI equation? Again, models are good for outcomes. Seeing a lot of interactions there, but when you start into the systems game, and when you add context and intelligence, you're executing differently and everything under the covers. How do you explain that to an enterprise? What's the strategy?
Deepak Khosla
>> Right, right. So two things. Number one, from the growth office perspective, I work very closely with partners. I'm going to answer your question in a minute, but it's very important to say I work very closely with our partners. So all the hyperscaler data product companies, we work very closely with them. Whatever we are building is kind of very tightly integrated what these guys are also kind of giving to their customers. And if you see, I did this analysis for one of our board meetings in April, the kind of new product launches that they have done, all the hyperscalers and the product company, if you see in the last six months is around context and memory. So they are also going that particular path and we are very happy. So now it's not about the discussion that we do. It's not about how do we go together. It's about the shared product roadmap. So that's a shared mindset that we have with them. So that's beautiful.
John Furrier
>> A lot of co-designing going on.
Deepak Khosla
>> Exactly. Exactly. Not co-designing with them, but the way we are building-
John Furrier
>> Co-collaboration.
Deepak Khosla
>> Co-collaboration. So now they know there's an impetus which understands this particular problem that our customers are having. They are building both their platform strength as well as services strength and they can basically understand our product well and implement that for the customer. To do what? Your previous question, two things. Either reduce the cost or increase the growth. That's the only two things which happens. Of course, customer loyalty is also very important. So eventually what we want to see is we want to see there's these agentic solutions which are there. Number one, they're not only the chatbots. Chatbots are good, great, very first unlock, needs to be done, but now are we thinking about use cases for manufacturing companies where we see their claims, merge them with the telemetry data, try to shift lift in terms of the production design side itself, this is what is happening. I'm sure Nachiket would have talked about the airline example, how do we figure out the various parameters that happens to make sure John's luggage reaches with John at the same time? How do we make sure that retail companies who want to basically do launches quite faster, the whole CDP process?
John Furrier
>> The airline example because, and I pointed out this to in a lot of conversation here, is that it used to be that you'd have to bend to the technology now that the technology bends to the use case.
Deepak Khosla
>> Right. Exactly.
John Furrier
>> And so there's no general purpose use cases anymore.
Deepak Khosla
>> Exactly.
John Furrier
>> They're all specific or tailor made or fit for purpose.
Deepak Khosla
>> Right.
John Furrier
>> That seems to be... Do you see that too with this?
Deepak Khosla
>> Yeah, absolutely.
John Furrier
>> All right. So what is the strategy for the engagement with the customer? Because a lot of enterprises are architecting a generational play here. So there's an entry strategy, there's problems to solve, but those problems have to be enabling to keep that headroom for growth because it's a bridge to the future that they're building. They want to have a 10-year horizon on the architecture. What's the pitch? I guess what's the value proposition when you come into enterprise say, are you solving the data problem first? Is it more holistic? What's the approach that you take with Leap?
Deepak Khosla
>> So the way we have set up, or I have set up, the whole offer strategy is we'll meet customers where they are. And in my mind at this point of time, they're at three places. Number one, customers are like, "All right, we did some experiments with gen AI and all that seems all right. We want to go big in, but we don't have that data estate. Can you help modernize, build our AI-ready data platform?" So over there, we'll kind of focus in terms of how do we bring AI, ontology, semantic thinking at the time of building the platform? A lot of my customers come for this whole modernization migration space, right?
John Furrier
>> On the data?
Deepak Khosla
>> On the data, building the data estate. Again, it's not about semantic data. It's not about the enterprise data. It's about the unstructured data and that's what is important. So number one, customers want and check with us, "Is my platform AI-ready? Or I want to basically launch this use case we are sold in terms of agentic customer service, agentic underwriting. What do I need to do in terms of having that data set?" So that's number one. We help them bring that context in while we are migrating and modernizing and building their data platform. Second is that my customer says, "I've invested a lot. My data platform is ready. I think I've got the right AI tools. I'm ready to launch good agentic solutions, solving good problems, not chatbots and all that, solving good problems in production at scale."
So that's where we'll come. We'll start building the knowledge layer, the semantic layer upfront. We'll not start thinking about it. Again, we'll basically bring the best of that enterprise knowledge to those agents. So that's the second point of interaction. Third, customers would say, "I've invested. I have almost everything, but my agents are not working. They're in production. They're not giving me the right kind of ROI." That's where, and most of the times, believe it or not, it's the lack of the context which is kind of stopping the agents to do what they need to do. LLMs are smart and they are very, very smart. But it's like you do a new hire, a new hire in the company, a smart person, and you don't tell them anything about the company, the smart person's not going to do anything well. But if you tell the smart hire about your businesses, your rules, and everything, the guy will do good. So that's the third point of incision, we make those agents smarter by bringing the context there in it.
John Furrier
>> Deepak, I love the use cases there and I think that's instructive. In fact, I've heard a variety of viewpoints, but I think it's general consensus in most senior leadership is, "Hey, let's go after the hard problems first because it's needle-moving." And so a lot of people want to go after not the easiest low-hanging fruit. "Hey, what's the low-hanging fruit?" "Oh yeah, we'll do a chatbot over here. Test it." I've seen companies get paralyzed by too many tests. They do too many low-hanging fruit use cases. I've seen a success where they go, "Hey, we know this use case and this workflow and the data set behind it actually creates revenue. Let's go there first." Because that's not like an IT project mentor. That's a CEO, CFO kind of approach. If that works there, then what do you need for budget? Okay. So take us through that because I think your second example pointed that because I think that's not a motion that you've seen in the past, unless it was a monster IT transformation project with massive budgets, big timetables. We've seen all those SAP deployments and the Oracles and all the big systems that have been rolled out in other generations now happening here, but different. Talk about what you guys do there because I think this is a sweet spot.
Deepak Khosla
>> It is. It is. And I think number one thing to do is to find out the biggest blocker that you have, the unlock that one needs to do. Imagine an oil and gas company which kind of moves their assets all along. It takes them about 80 days, multimillion dollar kind of a thing. It's a big, big area which needs to be unlocked. Imagine the whole underwriting process where you're going after customers. It takes you about five days, four days. Can you basically reduce that? Imagine an unlock in terms of a particular process which is very, very heavily human-oriented and can you basically make that little bit more automatic, autonomous, so to speak. So number one is the unlock that we basically go after. Second thing that we do is while that unlock would be complicated, the unlock would be big. We try to break it down. We are big believer of the skill concept. So we want to break down each particular scenario agent into kits first and kits right below the skill level and then start building those skills. Those skills, let's say even for a fraud use case, which is big use case, could be to figure out what is false positives to a particular human-in-the-loop? So just a skill which can kind of figure that particular thing out. So start building those skills and those skills will turn into kits, those kits are already deployable, and those kits then you can engage in an agent ,and then you can think how much level of agency you want to give to that particular agent to be autonomous or not. So that's how you kind of break now the bigger problem and still you're attacking the bigger problem. You're still bringing those CEDL lifecycle methodology that I talked about and you try to go to production.
John Furrier
>> I just wrote a post, Substack, on LinkedIn and SiliconANGLE on the term I call hyperscaler 3.0. And we're kind of in 2.0 now, which is enterprise adoption, neoclouds. 3.0 is more of a frame around what we see happening. And I want to get your reaction to this because you hit on some of this, which is you have a fully distributed hybrid cloud infrastructure with now an angentic layer that's infrastructure with AI-native agents and builders on top building value. So I wrote the blueprint for the next phase of the enterprise is architecture, receipts, which is proof, and outcomes. What's your reaction to that?
Deepak Khosla
>> Oh, absolutely. I think the kind of investment that's going in, I think there would be a come-to-Jesus moment that, "Show me the output. I mean, all the tokens that I'm spending, what's going to come out of it?" So definitely I think that's what I think 3.0 is going to be, which is more in terms of outcome, more in terms of ROI. I believe that ROI would be to start with more in terms of perhaps efficiency, growing markets. If you get the right context, it's about having that poll position in terms of the new world that you are in. There would be a lot of consolidations. You want to be the one who does the consolidation and who shouldn't get consolidated and I think AI is going to help all that.
John Furrier
>> Okay. So how would you describe, the final question for you, Impetus Technologies' blueprint for the enterprise, what's your pitch?
Deepak Khosla
>> I think leap.ai, we come, we help your agents work better. We help build the enterprise context for your agents that you have. We bring a platform approach, we bring our FDE approach, and we work with hyperscalers. We work with your landscape that's there. So time to market.
John Furrier
>> Time to market is your hyperscaler 3.0 solution.
Deepak Khosla
>> Yeah.
John Furrier
>> This is where it's at. Well, congratulations. I love the timing of the launch. It just wasn't invented to fit the fashion of the market. You guys have done the work with the services with the beachhead. You've been in the front lines getting those requirements, building a trajectory. Let's see how it goes and wish you the best of success.
Deepak Khosla
>> Thank you so much, John.
John Furrier
>> Congratulations.
Deepak Khosla
>> Thank you.
John Furrier
>> I'm Jean Furrier here. The launch of Impetus Technologies Leap AI. This is the leaping of the next generation. Again, the architecture's changing. It's evolving. Architecture-proof receipts and the outcomes ultimately is what people are focused on. Apply the AI to where the money's being made, where the ROI is. That's where the focus is. A little bit different than the IT projects. Obviously, the CFOs involve a lot more of the C-suite. We're doing our part here in theCUBE to bring that to you from the NYSE Wired program. Thanks for watching.
John Furrier