Episode Transcript
[00:00:03] Speaker A: Welcome to INA Insights where prominent leaders and influencers shaping the industrial and industrial technology sector discuss topics that are critical for executives, boards and investors. INA Insights is brought to you by Aina AI, a firm focused on working with industrial companies to make them unrivaled segment of One leaders. To learn more about Aina AI, please visit our website at www.aina.AI.
[00:00:40] Speaker B: Hi everyone and welcome to another episode of our Aina Insights podcast where we continue our focus on early stage AI companies. Today our guest is Mr. Vibhanshu Abhishek who is the CEO and founder of Alts AI. Alts is a cutting edge generative AI platform designed to build skillful and accurate AI assistants. These AI assistants can be a game changer for sales and support teams, helping to close deals faster and reduce support costs significantly. Along with Altius, Vibhanshu is also working with University of California, Irvine as an Associate professor of Information Systems and a charter member of Tai Socal. He is also an advisory board member of Brevi. Vibhanshu comes with more than 25 years of experience in the field of engineering. Vibhanshu, a very warm welcome and very excited today to talk to you about Altius and your own journey as its founder and CEO.
[00:01:36] Speaker C: Thank you so much Nidhi. Really wonderful to be here. Thanks for the opportunity and looking forward to talking to your audience and speaking with you on some of the topics we discussed earlier.
[00:01:46] Speaker B: Awesome. So why don't we start with talking about Alts Vibanchu and if you could tell us why, how did you start the company?
[00:01:54] Speaker C: That's an interesting, you know, I'll take you down a trip memory lane. I've been doing AI for probably over 25 years now, so most of my adult life and then some, I would say really been fascinated by robots and artificial intelligence and machine learning for a long time and I was trained as a computer scientist, went back to grad school at Wharton where I was one of the first people who started applying machine learning to business research and then ended up being a faculty at Carnegie Mellon which had the first machine learning department in the world.
And more recently I've been on the faculty at UC Irvine. So the reason I sort of tell that story is I've been someone who's been very excited with AI for several years and with generative AI I really started seeing the potential and really the step change that this technology could bring about. Right. And this is sort of from an insider perspective, this is not sort of taking an outside in view. This is really sort of Being someone who has been building these AI models and I could see the step change in what generative AI could do. So really started Altius as a way to bring generative AI to enterprise use cases. And then over the last couple of years, and this was, you know, way before ChatGPT happened, maybe like a year before ChatGPT happened. So we were a little bit early to the party. And then over the last year or so we have gravitated towards looking at sales and support use cases, primarily in the BFSI domain. But you know, can be. Technology can be applied broadly, but that's how we are approaching our customers and our use cases.
[00:03:34] Speaker B: Got it, Got it. And as you're doing that, what in your perspective sets Altius apart from some of the other companies?
Because especially now, generative AI is like so popular. There are so many who are working in this space. So what according to you sets altars apart from them?
[00:03:53] Speaker C: That's a very interesting question. So, you know, I think there are a few different factors that set us apart, right? One is a whole philosophy of how we are approaching this problem. The first thing that we are looking at is if you look at customer centric functions, so marketing, sales and support, they have been very siloed for a very long time. And that was essentially because it was people who used to lead these functions. So people are limited in the capability. Hence software that was developed to help these people were also limited in the capability and very siloed. So I think the first thing that we're fundamentally doing is breaking these silos and saying, how can we build an AI first layer for customer interactions? It can be aided by humans or it can be completely autonomous, but really how do we build an AI first worldview of customer interactions? And I think that really sets us apart and philosophically how we think about building and architecting the platform where we're pulling in all these structured and unstructured information for customers, for companies, so that we can have a very deep conversation with customers that we could not have had earlier across different parts of their buying journey. You know, given that philosophy, what we have done is we have built a platform where our customers, which are large, mid to large enterprises, can go in and pick pieces to create any kind of customer journey that they want. Right. Or customer experience that they want. So that can be applied to customer support, it can be applied to lead generation, it can be applied to sales enablement, or it can be applied to assisting the internal call center or support teams. So it's very flexible in that format. And Very extensible. But also it's powered by a lot of our core technology that we have built like that I've built particularly over the last two decades that looks at how do you bring a level of accuracy and determinism to generative AI. So generative AI is great like ChatGPT is great at. We give it something, it'll spit out an answer for you and that answer mostly looks correct in how the answer is generated. But it might not be very accurate or it might not behave deterministically every time that you ask something offered, which is what need in a lot of domains that we are going after. Like if you think about insurance or banking or fintech, you want to be very accurate in the answers you give. So we have built the symbolic AI layer that brings in this level of determinism and accuracy that is just not possible with large language models. So we have no hallucinations in the platform.
Salesforce is still talking about low hallucination. We have no hallucination for the last two years since we have been in production. So I think that sets us apart. We are also deployed at scale. Many companies are still doing POCs. We are deployed at scale delivering real value to customers and I think at the end of it all we are just powered by a very high powered team.
I've been at the forefront of this area for many years, I've taught many students in this area. So we're really subject matter experts in this area. So you know we have been able to fortunate to build a team that understands this technology in and out and so that's why we are able to deliver this value to customers.
[00:07:13] Speaker B: Yeah, that sounds fantastic. And Vibhanshu, in terms of like the function it is clear the focus is sales and sales support.
What are some of the end markets where you said right you are like at scale at some customers and you are also like doing POCs. So what are some of these end markets that you're serving at the moment or looking to serve in the future?
[00:07:35] Speaker C: Yeah, I would say two mainstays for us. One is FinTech, large FinTechs that are scaling up dramatically. So Angel1 for example is a big customer of ours. Insurance is another area that we are making a lot of strides. So bunch of insurance companies both across us and India, customers like company Prudential Compass in India Acco and we are in discussions with a bunch of other very well known brands doing PoCs. So I would say these are two big areas that we have a lot of momentum. We are also talking to several banks. So we have done wrapped up POCs with DBs development bank of Singapore. We have been doing PoCs with some community banks in the US, also in India. So, you know, bunch of things going on, but really looking at use cases that are very high volume, that require a high deal of accuracy, but also industries where it's complex, products are complex, use cases are complex, customer needs are very heterogeneous. Those are the areas where I think generative AI has the maximum value to offer. And so that's the area that we are going after.
[00:08:43] Speaker B: Right, and you said that you've been in production for about two years now. So as you've been doing these projects with customers in these end markets, what sort of a feedback loop do you have with them as you're building solutions with them or for them?
[00:08:59] Speaker C: Yeah, so, you know, that's a very interesting question that the challenge is AI is very new and especially generative AI is even newer. So no people, if you think of buying CRM software, people know what exactly they're buying, but when they're buying into AI, it's not very clear. So there's a little bit of education to what can be done and what cannot be done. And you know, that's where I like to bring my academic hat on and sort of tell people, you know, this is something that like instead of promising the world and not being able to deliver, really educating people about what is the state of the art and what can be done and what cannot be done with AI, I think is probably the most important thing to do. But secondly, I think what the market is looking for is really solutions to their problem. They're not looking for packet software, they're looking for here is a problem that I have. How can you solve this using AI? And for the most part they don't even care whether it's AI or not. But really, can you solve that problem? So typically we work with large customers trying to understand what is it that they want to do and how to go about solving their problem. What helps us, as opposed to many other firms, is we have this platform approach. Almost think of us as we provide the green base plate for Lego and then we have all the pieces which can be configured very quickly to solve a specific problem. So our go to market, we are very extensible and we can adapt to what a particular customer's need might be and what fits their problem and hence a solution for them, but also something that can help us take the solution to market very quickly. Right. In production, in a matter of weeks instead of several months.
[00:10:43] Speaker B: Got it. And any interesting use cases, Vibhanshu, that you could elaborate on for the benefit of our audience?
[00:10:52] Speaker C: Yeah, you know, happy to do that. So, you know, Angel 1, for example, is a very, you know, one of the leading brokerages in India, probably the second or third in volume. We started working with them with about 15 million customers. They're probably about not 23 million customers. So they've grown dramatically in the last seven, eight months that we have worked with them. We are handling like 60, 70% of their tickets. So they've been really being able to scale without increasing their support costs dramatically. In fact, we have been able to reduce the support cost to some meaningful extent. So, you know, think about this in terms of outcomes. We are solving hundreds of thousands of tickets for them, getting closer to millions of tickets for them every month, saving them millions of dollars yearly. So really meaningful outcomes for a publicly dated large company. We have seen very similar outcomes for some other bigger companies. Prudential on the other side is, and this is something that we're seeing with some other insurance customers, is helping their sales team sell better. So they've been able to reduce their training times dramatically by 60, 70, 80% in some cases. And also these agents, call center agents, are now able to sell more. So, you know, you're reducing the cost base plus also increasing the top line. So again, huge roi. And that's sort of where I see the impact of generative AI. A lot in the industry is in sales and support use cases where we can have dramatic impact on both the top line and the bottom line of the company.
[00:12:30] Speaker B: Got it. And Vimanju at ina, we focus on the industrial sector. I know you mentioned that most of your customers today are in fintech, banking and insurance. But based on your experience, not just at all tiers, you've been grappling with AI for almost 25 years. Any useful insights for the sector on where they could apply AI at?
[00:12:52] Speaker C: Yes. So, you know, actually we have a partner that. So we don't go directly into infra and manufacturing, but we have other customer. We have a customer called Tacit that is using our platform to go into that vertical. And essentially there are many use cases of generative AI. So think about, you know, a lot of the equipment. There's a lot of know how required around how to service these equipments. You have big field tech teams who go and solve problems and a lot of times, especially for, you know, large companies like Adnoc or, you know, big oil and gas companies, you have people are moving around with manuals trying to figure out, you know, what does that error code mean? And so we can train these generative AI assistants very quickly on all of the product manuals, documentation, SOPs, and make it available to these field agents very quickly. So something that would have taken them 15, 20, 30 minutes to solve, they can now solve in a matter of seconds because they have all the information that comes to them. Right. And think of being on an oil rig. Every time, every second that you're not solving that problem, you're losing millions of dollars in cost. Right. So. So I think those are definitely areas where generative AI can be applied. AI can broadly be applied in predictive maintenance. I've done a lot of work with many large customers on how can we build predictive AI models for saying no, how can we do predictive maintenance, for example, to reduce costs, to improve efficiency? So there are many applications of that, of AI in this domain.
[00:14:26] Speaker B: Yeah, yeah, understood. And Manchu would just like to go back what you said in the beginning that you've been in this field for 25 years, and according to you, generative AI is definitely a step change in the advancement of the technology. Right. So what would be some of the other pivotal moments like that that you have observed in the 25 years of your career in AI?
[00:14:51] Speaker C: So, you know, I think I would say again, I just want to situate this. This is the third wave of AI, right? So AI is not new. It's been around for the last 70, 80 years now. And so we have gone through the first wave of AI, then we went to the second wave of AI. So I'll not get into the details of that. But now I think we are living in the third wave of AI. And there are two before generation happened. I think there are two dramatic things that happen. One is there was both data and compute that was massively available. You know, beginning of the century, you know, we had all this data on the Internet. People were producing social media data and whatnot. Right. We were produced as individuals, as a race. We have just the amount of data that we're producing has exploded, which did not happen in the last century. Second thing that happened simultaneously was the compute cost and the storage cost went down dramatically. So with, you know, AI and specifically with machine learning, the problem was always there wasn't enough data and there wasn't enough compute. So when I started working in this area, it would take a long time to build even a small neural network model. Forget about a trillion parameter model, even a hundred parameter model would take a few days to train because compute was so slow. And now with GPUs and whatnot, the training cost has reduced dramatically. So I think the first wave of change came about, the breakthrough came about because you had a lot of data and very cheap compute and storage. Second came about with deep learning and imagenet. So with the work that Jeffrey Hinton had been doing for the last three, four decades now, he could apply with all of this new data and new compute to build deep learning models. And I would say that was the second breakthrough that happened in this third wave of AI. And I think the third is the transformer models, like the attention paper. And of course everything that's happening with generative AI models now. So I would say that's probably the third breakthrough that has happened in generative AI that we are getting closer to human level intelligence.
We are seeing it now. It's not happened overnight, it's taken like 50, 60 years. Jeff Hinton was working on deep learning for probably 30, 40 years before, you know, he saw the light of the day and he had compute and data to be able to go and prove what he had been working on.
[00:17:17] Speaker B: Yeah, but it does seem to people that generative AI was born overnight when ChatGPT came into being. And 2023 was like a highlight here. Right, for generative AI. So, but what's your take? I mean, of course there's been a lot of progress like you talked about in terms of both the infrastructure and also the use cases that we are seeing being deployed. But from your perspective, what's your take on the real progress in the field of AI versus just being a mere hype?
[00:17:49] Speaker C: You know, I would say that there is again, you know, there is hype and I think it's lack of education really that drives that hype. So in terms of impact, I see tremendous potential. But you have to apply all of these techniques to the right problems, you know. So as an educator for many years, I've taught about how you should think about AI projects. Should you even apply AI to many projects? Many times AI is this catch all thing that say, oh, it's almost like magic. Right here is a problem, let's apply AI to solve this and it doesn't work. There is science to this, there's a lot of art, but there's also a lot of science to how you apply AI. So I think the first order of business is are we even applying AI to the right problems? And if we are, I think there is tremendous potential. Right. So, you know, some of the Use cases that we are going after this. Some things could not have been done earlier and now you can see dramatic changes in how those things functions can change. You can probably run a call center with probably one tenth the number of people with probably better outcomes than what you could have done earlier. But then there are many areas where we don't know whether the outcomes will come or not, whether ROI will come or not. So we have to be very meaningful about where to apply these technologies. Second thing, I think we have to be very meaningful about what's the shortcoming of these technologies. Right. And so how you design systems, you have to be careful about what's the limitation of them. And then how do you overcome the limitation either by having human in the loop or having checks and balances in place to make sure that these systems behave in a way that you want them to behave. Right. I think the big thing about AI versus other types of models are software program, for example, is software programs are very deterministic. You know exactly what they're going to do. But with AI in general, the challenge is it keeps learning on data. And so as you get more and more data, the model starts performing differently than what it was programmed for. And so you have to keep constantly assessing the performance of these models. And the generative AI, it takes it to an extreme where it's probabilistic in nature. Right. So I think you need to be aware of all those facets when you're deploying them in production and making sure that they will evolve over time and hopefully they're evolving in the right direction, but also making sure that you're experimenting and checking if things are evolved where they need to be.
[00:20:22] Speaker B: Got it. And on that note, Vibanshu, last question.
How do you then foresee the trajectory of, let's say, generative AI in particular? Right. Because that's the area that you focus on, especially at all tiers.
[00:20:37] Speaker C: Yeah.
[00:20:37] Speaker B: And how do you see adoption panning out?
[00:20:41] Speaker C: You know, I think that's a. So I think the very long term view is probably everything that we know today is going to change. How, what kind of work we do, who does the work. All of that is going to be different in the future. I think it's going to be very, you know, difficult for me and probably, you know, I don't want to even make that comment on how it's going to change. I think we are all going to wrap our heads around how things are going to change. It will be a big societal. It's not only going to be a big technological change, but also a big societal change 10 years from now. But I do see increasingly there is a lot of demand for applying generative AI. I also see a lot of value in applying generative AI just based on what we have seen with a lot of our customers. So increasingly there'll be more and more adoption in the medium term. You know, I can definitely say in the next three, four years, a lot of companies will have workflows that have generative AI in those workflows, one way or the other.
There might be, you know, there might be this stuff of disillusionment in the short term of people not realizing the value that they're seeing. I think people are already talking about Microsoft Copilot not being very useful. You know, I hope Microsoft doesn't come after me for saying this, but I think, you know, some of the customers that we have been talking to have been saying that, you know, it's not as useful as they expected it to be. And I think that's something that we have to be very careful and cognizant about that, you know, there's no general purpose AI today that solves all problems. We have to still think about developing solutions for specific needs, specific use cases. And that requires just more than having an LLM solve all your questions, right? You have to design systems to solve real problems. And that takes understanding, that takes effort, that takes diligence, which is not just, hey, can you make an OpenAI call and solve all the problems in the world? So I think that's where I would sort of leave it at right now.
[00:22:38] Speaker B: Awesome. Well, thank you so much, Vibanchu, for making time for us today and sharing your perspectives on a very hot topic, generative AI. Really appreciate this.
[00:22:49] Speaker C: Thank you so much, Nidhi. This was really wonderful.
[00:22:52] Speaker B: Thank you.
[00:22:58] Speaker A: Thanks for listening to INA Insights. Please visit INA AI for more podcasts, publications and events on developments shaping the industrial and industrial technology sector.
