Autonomous AI Systems: Entering A New Era Of Technology With Shirish Nimgaonkar
20:23 Share

They become very good at what they do, but only 0.1%.

So I went to college to study medicine.

engineering, in mechanical engineering, but predominantly focused on math and what used to be called at that time as operations research, which in many ways is the precursor to the modern artificial intelligence frameworks. Are those the expert systems or something? What did you call it? Operations research. It was just, you know, a fancy word to talk about

a whole range of optimization problems. It's essentially a combination of statistics and analytics. And then subsequently, we sort of induced big data in the picture as well.

I worked a lot with thinking about complex systems, figuring out ways to quantify them, and then come up with equations that would help us to construct complex systems and essentially help us to solve them and also predict what can happen with respect to the changes in those systems over a period of time. And so that's what I'd studied. I went to a college called Indian Institute of Technology, which is one

one of the very competitive colleges in India. Graduated then and came straight to Silicon Valley. I was always very excited and fascinated in terms of what was going on. That was during the dot-com boom. So a lot of exciting activity happening with respect to Netscape and Yahoo and the birth of Google as well. So I did my master's at Stanford.

again, focused more on operations research and industrial engineering, thinking about large systems, ways to quantify them, and again, ways to think about what might be ways to predict what can happen in the future and hence proactively address the issues and actually solve the issues as well. So,

Tell me about Bliss AI then. What was the idea for it and what is it? Essentially, Bliss was born based on several years of my experience. If you look at the problem statement essentially in the world right now, is we have, you're surrounded by a whole range of devices. You have your laptop, cell phone, tablets as one category of devices. Then on the factory floor, you might have other categories of devices. For instance, your robotic terminals,

or even your car is essentially a device now, if you look at a Tesla. And then in the home environment, you have, you know, your smart microwaves, your dishwashers, your washing machines and all, right? So there's a whole range of devices, different kinds of devices, and also distributed geographies for those devices. But the support model in terms of device management is still very, very old school. Essentially, you wait for your laptop to crash,

because Outlook is not working, or you just wait for your dishwasher screen to crash. And then you don't really understand what the problem is. And so you call a technician or an IT or help desk, and they come in, they take a look at what might be wrong in your device. They try to go through some heuristic solutions. And sometimes it works, sometimes it doesn't work. And then they come up with a fix and fix.

you know, when it works, you hope that the fix stays on. So essentially the problem statement is that it is, the reaction is slow and delayed, number one. Number two, it's generic and impersonal, which means it's not focused on actual problem for that particular device.

And number three, in many cases, you don't really understand what is the root cause of the problem. So even though you might come up with a solution, you don't really know whether it's going to work and sustain over the long term. Right. So the inverse of that is you essentially need a platform or a solution that is predictive, which can actually figure out when the problem is going to happen before the problem actually happens. Number one. Number two, it can come up with a solution immediately.

which is personalized for that particular device or that particular end user. And number three, it actually understands what the specific nature of the problem is and can come up with the solution to solve that particular problem. And all this by self-learning over a period of time. And that's essentially what we do at Ableist AI, which is it's an autonomous self-learning platform that

that can predict when the issues are going to happen in different kinds of end-user devices. What's a specific example of what it's done? For instance, let's say your Outlook is not starting. So what would you do at that point of time? Either you download some kind of a patch or you restart your device, or you might just call IT and Help Desk and they might just come and do a bunch of things on your laptop and then

You know, hopefully it starts working. But there are actually 25 reasons as to why your Outlook might not really start.

And many times we've seen, even though IT might come and fix your issue with respect to Outlook not starting, the issue could likely reemerge. As we have seen, it does reemerge in almost 77% of the cases within a short period of time. So you haven't really solved the exact nature of the problem. So what Eblis AI can do is it can actually figure out, it can predict when the fault is going to happen over a period of time.

because we have these complex models that understand all the different contributing factors, it can identify the exact nature of the

a problem with respect to which of the 25 reasons exist in this case as to why Outlook is not starting. And it can automatically trigger the fix for Outlook not starting even before you recognize that the problem is going to occur. So hence, what's the value? You have saved on the cost of ID support, right? A, you have saved on your loss of productivity because the device is available and the end user or you are more productive. And

And number three, you just have a much better experience because of all this. Okay, but I mean, has it been able to fix things like, you know, third-party software? Does it need an API or how does it?

Exactly. So what we do is we have our agents that go and sit on different kinds of endpoints, what we call as different kinds of computing devices. We actually gather a lot of data from all these different devices. So we have five or six different kinds of data that we would typically gather, which are a combination of the endpoint data, the service tickets that

are typically filed by the end users for the complaints. Then we have a lot of customer data as well for the deployments that we have. Then we have our own proprietary chat that we engage in to understand the problem statement a little bit better. Then we have the public data and we use a lot of synthetic kind of data as well. Combination of all this is what leads into our

proprietary AI models where we use deep learning methods. We use our small language models that we train based on our understanding of this particular domain. And we use a lot of agentic AI as well for us to engage in reasoning and better understanding and actually understand

attempting the solution, right? So all this leads into our self-learning knowledge base, which is where we have a combination of all the different kinds of specific issues and then the specific solutions that we trigger for these kinds of issues.

Let's take a look at 4N and after you can tell me about on what metrics were improved or what was, you know, what changed once they used Eblis AI. So typically the metrics are along three dimensions, right? It's usually it's cheaper, faster and better. We have shown that we can actually cut down on the cost of operations from anywhere between 40 and 70 percent. We can decrease the loss of productivity between 45 percent and 75 percent. And then in that

all that leads into the ROI, which is typically in the range of 15x to 45x. We have to know that as a comparison, most of the existing solutions out there, they typically lend themselves to ROI, which are in the range of 3x to 5x. And the reason why we are significantly better than most of them is because we can enable significantly more automation in terms of the percentage. So we have shown that we can actually solve predictably

anywhere from 25 to about 65% of the issues, in some cases even higher, depending on the nature of the issues without any kind of human intervention. And above all, all that leads to significantly better end user customer experience. So what problems does the AI seem to be more amenable to and which ones are tougher for it?

Before we continue, I've been personally funding the Finding Genius Podcast for four and a half years now, which has led to 2,700 plus interviews of clinicians, researchers, scientists, CEOs, and other amazing people who are working to advance science and improve our lives and our world. Even though this podcast gets 100,000 plus downloads a month, we need your help to reach hundreds of thousands more worldwide. Please visit findinggeniuspodcast.com and click on support us. We have three levels of membership from

from $10 to $49 a month, including perks such as the ability to see ahead in our interview calendar and ask questions of upcoming guests, transcripts of podcasts you're interested in, the ability to request specific topics or guests, and more. Visit FindingGeniusPodcast.com and click support us today. Now back to the show.

So we have different categories of problems that we go after. You can have application-level problems. You can have issues from an operating system standpoint. You can have

you know, certain kinds of connectivity problems and so on and so forth. So generally we would classify the problems along these different types, and then we would also classify them based on their complexity. So typically we have been able to solve most level one problems, which are sort of more simplistic problems and a significant percentage of level two problems as well, which require more detailed understanding from a domain standpoint and also more complex problem solving.

And hence, it's more challenging for us to sort of execute on those kinds of solutions as well. For instance, you know, restarting an Outlook, that is level one, right? But then figuring out why Outlook is not starting and sort of going through the reasoning process, categorizing and coming up with sort of specific tailored solutions, that is usually a level two task. And we have been able to do not only most of level one, but a significant chunk of level two tasks as well. And that

sort of extends across all the different categories of issues that I outlined earlier. Is this only for large enterprise or is it for small business? Like what kind of pricing and models do you have available? Right. So this solution is within the realm of a new category of solutions called autonomous end-user computing. And currently we are focused on very large enterprises, the likes of Fortune 2000 companies that

that have a huge volume of devices and also different varieties of devices in addition to the distributed nature for those devices. From a conceptual and technical standpoint, this solution can be applied to any computing device. And so the second area of focus for us is the managed services providers who actually service small and mid-sized enterprises as well. At

At some point, we will be going to the end users or the individual subscribers too. But just from a go-to-market strategy standpoint, our first focus is the large enterprises followed by the managed services providers that focus on the small to mid-sized enterprises. Usually from a pricing standpoint, there are different ways to think about what might be the best way to capture value. But a simplistic way would be the pricing, which is based on a per device per ecosystem.

year kind of a model. But the kind of deployments that we go through, these typically tend to be for several thousand devices and the contracts tend to be for multiple years. Okay. Are there certain applications that, I don't know, are hardened somehow that you can't access or interact with? Or are there ones that you can only interact with kind of from like a user's point of view, there's no way to get into the code or really do anything with them? So we work very closely with the different kinds of enterprises,

that are interested in this solution. So if you think about the complexity of the solution along with the potential for the solution going wrong, which is sort of the risk of the solution, essentially, we want to focus on

Certainly on adding more value and hence going up the curve with respect to the kinds of complexities that we are dealing with. At the same time, we do want to minimize the risk as well in terms of something going wrong from an automation standpoint. And hence, we work very closely with the users, especially the large companies, in terms of getting their approvals for the kinds of problems that we're dealing with and also the kinds of fixes that we believe we can trigger and

and hence automate the solutions for. And so to that end, a lot of our solutions that we trigger do have the provision of asking for approvals from the end users and also from the technology admins as well in case they're not comfortable with our automated triggering of solutions that we might just ask for their approval to trigger

essentially go through the process. So that takes care of the question that you just asked, which is if they want us to exclude certain kinds of applications or certain kinds of issues that they deem as high risk from an automation standpoint, that can certainly be done. In terms of industries that are interested or that see the value of this, which ones seem to be most amenable and which ones are like, even though they need it, they're resistant to it? We have seen that the interest and the response from

for the E-Place solution has been phenomenal across a wide range of industries. So we have seen strong interest in deployment from retail, certainly a lot from manufacturing, a lot from oil and gas, even aviation, financial services as well. Of course, there are certain industries

industry regulatory issues that we have to deal with in terms of data security and compliance and all that. And that tends to be more in certain industries like healthcare and financial services compared to some others. Having, keeping that aside, I think because the value proposition of the solution is so compelling, it

it sort of overcomes any initial constraints or hindrances that we might have from an adoption standpoint because of some constraints from a regulatory standpoint that we might need to go through. What are the industries that are tending to use the product most right now? So we've seen a lot of interest from retail, from manufacturing, from consumer products, from aviation as well. So think about it. Any industry that has a huge range of devices,

in terms of volume, variety, and also distributed geographies for those devices, essentially needs this kind of a solution that is predictive, proactive, personalized, and can automate the detection, diagnosis, and remediation of any kinds of issues. So the autonomous nature of the platform is what makes it truly compelling. And you've heard about autonomous

autonomy across a different variety of industries, for instance, self-driving cars is a great example of what you would call as autonomous driving. Similarly, the next generation of computing platforms are largely going to be autonomous. And that's where you see the emergence of classical deep learning, agentic AI, which we're starting to hear a lot more of, which we use quite a bit as well in our platforms. So

Essentially, the value prop is very compelling across various set of industries. But to your point, the initial traction has come from industries which tend to have lower regulatory hurdles and also huge volume and variety of devices across distributed geographies. Is there any legislation surrounding AI or is it way too early? Anything impacting you guys or that you see coming or even see a need for that's not here?

We are in the early stages of AI adoption. There was a very interesting article in the Wall Street Journal about how there has been a mandate across a variety of companies to think about AI adoption. However, there is no

the value that the companies have been able to derive from the application of AI has been limited at this point of time. And a lot of that has to do with how the AI is sort of structured more from a governance standpoint and how the value is executed on from an organization standpoint. And while

The frameworks for AI governance are still evolving. What we have seen in terms of AI adoption and where we have seen compelling value being obtained is where you have strong buy-in from the senior management. The focus is really on business objectives as opposed to a horizontal application of AI and where there is

strong commitment from the organization towards execution of certain kinds of solutions that help them to achieve their objectives. In our case, we work very closely with the CXOs of different kinds of organizations. We clearly have an understanding and work with them in terms of developing a shared understanding of what their problem areas are from a device management standpoint and

And then we come up with a platform and then we try to tailor the platform based on what is specific in terms of the nature of issues that we have seen there and then have clear roadmaps in place to articulate and execute and the value that we are projecting because of our platform. And all that leads to quantifiable metrics and hence the ROI that they can see over a period of time. Very good.

Well, what's the best way for people to find out more about Eblis and see if they qualify to use it? Well, you can certainly visit our website, eblisai.com, and you can feel free to send me an email. It's just shirish.mangaukar at eblisai.com, and we would be delighted to engage with you to figure out if we are the right fit and what might be the best way to create a compelling value for both the parties. Well, very good. Thank you for coming on, Shrish. I really appreciate it. Thanks so much, Shrish, for inviting me.

If you like this podcast, please click the link in the description to subscribe and review us on iTunes. You've been listening to the Finding Genius Podcast with Richard Jacobs.

If you like what you hear, be sure to review and subscribe to the Finding Genius Podcast on iTunes or wherever you listen to podcasts. And want to be smarter than everybody else? Become a premium member at FindingGeniusPodcast.com. This podcast is for information only. No advice of any kind is being given. Any action you take or don't take as a result of listening is your sole responsibility. Consult professionals when advice is needed.