Earley AI Podcast – Episode 58: AI-Driven Manufacturing and Data Transformation with Sahitya Senapathy

Breaking Through Chatbot Myths: How Manufacturers Drive Real ROI with Generative AI and Data Standardization

Podcast Transcript: Overcoming AI Misconceptions and Data Challenges in Industrial Manufacturing

Transcript introduction

This transcript captures a conversation between Seth Earley, Chris Featherstone, and Sahitya Senapathy about the critical gap between AI aspirations and manufacturing realities, exploring how organizations can move beyond chatbot misconceptions to implement data-driven AI solutions that deliver measurable ROI through workflow automation, data standardization, and advanced analytics while addressing data privacy concerns.

Transcript

Seth Earley: Good morning, good afternoon, good evening. Welcome to the Early AI Podcast. My name is Seth Early. And I'm Chris Featherstone. And today we're really excited to introduce our guest to discuss industrial products, industrial manufacturing, product data, product information management, and how these types of approaches can be used. AI approaches can be used in that particular space, especially when we talk about improving product data. So we'll talk about use cases for that, talk about generative AI and data privacy, and we're also going to be talking about solution, execution and deployment. So our guest today is an expert and visionary in the field of AI and industrial manufacturing. He's currently serving as the CEO of Endeavor. He has a remarkable career that includes both working with prestigious institutions such as the US Air Force and studying at the University of Pennsylvania and the Wharton School of Business. His professional journey also included stints at AWS and Palantar, where he honed his skills in leveraging advanced technologies to solve complex problems. Sahetia and I didn't ask for your last names. You got it.

Welcome to our show. Nice to have you.

Sahitya Senapathy: Thanks for having me. Really excited for this.

Seth Earley: Terrific. So, you know, we like to start off by trying to understand what are you seeing in terms of misunderstanding, especially in this industrial product space? What is it that people are not getting or that you have to go in and explain? What are some of the misconceptions about AI, generative AI and industrial data.

Sahitya Senapathy: Wow, Getting right into the meat. I like it. You know, I think the really, the really interesting thing is, you know, folks just aren't, aren't familiar with what generative AI is. And what happens is this causes a bit of confusion because people think chat bots are the panacea. You know, they think that chatbots is going to solve all their problems. And so when we walk in and people are talking about use cases, people are like, hey, should we, you know, be talking to a maintenance manual? Right? And when we think about, you know, why we actually want to implement generative AI at the end of the day in the industrial space, it really comes down to roi. How are you driving top line sales? Or how are you saving me costs and reducing my opex? And the reality is that chat bots aren't going to get you there. And that's usually the first misconception that we have to handle. How can we use generative AI to drive real ROI and real use cases?

Seth Earley: And so how do you go about that? Where do you start when you go into an organization? And so they think that the chatbot is the answer but obviously there's a lot of stuff that needs to be done. It's conversational interaction with their data, using their own data, as opposed to the LLM itself or the generative AI itself. But where do you begin that process when you start talking to companies?

Sahitya Senapathy: Yeah, really good question. Because when you're dealing with enterprises, it's often the case that you have a bunch of different stakeholders. And so you have know, I'll, I'll put it into two broad buckets of people who are going to be vital to this process. The first is IT information, right. They're going to be the ones that have the governance and understanding of the data that surround the organization. And the second is the business case user, whether we're talking about the sales department or the operations department, the supply chain department that's in charge of procurement. And you need both of these forces to come together to come to an understanding. Because to really solve these problems in enterprise manufacturing, you're dealing with issues around siloed data. You're dealing with issues around messy, unclean data sources and different ERP systems. And what this means is that not only are you feeling the immense pain from a business use case standpoint, but you're also seeing the effects in the IT department. They're struggling too. And so our process always starts with discovery, with IT and with the business case users to identify, you know, where are you feeling the pain in your business? Where can we improve your margins? But what does the kind of data layer look like that is creating this issue for you?

Seth Earley: And do you want to provide maybe a specific example of where you, where you will find those kinds of opportunities in a manufacturer? And of course, industrial manufacturers many times are not at the leading edge of technology. They're spending more time on manufacturing processes and making investments in their operations. And many times IT and technical capabilities lag a bit. So tell me a little bit more about what you see on the ground and what those specific needs might be.

Sahitya Senapathy: Yeah, so I'll kind of give you the IT standpoint and how this ends up manifesting in the business use case with the IT folks. You know, you have this accumulation of legacy softwares in the industry. So you've got ERP systems, right? Your enterprise resource planning like Oracle and SAP, then you've got the manufacturing execution systems, the meses, and then you introduce all sorts of different unstructured data sources. And this is actually the long tail of unstructured data is really where generative AI comes into play as a massive infrastructure force multiplier for these companies. And you've got email, you've got spreadsheets, you've got Word documents, you've got PDFs, and every single manufacturer, whether you're talking about small mom and pop shop that's on the street across the corner, or you're talking about, you know, the biggest manufacturer with 30 plants across the world, they have this problem where different plants that have, you know, come together as a result of M and A activity have different ERP systems, they have different MES systems, they're collecting different sensor data on the shop floor. Every person at the plant's got their own spreadsheets that they're doing analysis and Word documents. And the first step is kind of just analyzing this. And, you know, I'll give you a tangible example. Let's say we're dealing with sales folks, right? And we've identified, hey, you know, there's a problem with a lagging sales volume for the last five years. We want to start automating our sales workflows. What this means is we started looking on a plant by plant basis, and the results might be, hey, you have different product catalogs in different ERP systems. One plant might be on SAP, the other one's on Oracle, and there are discrepancies in the product catalog, and there's unreconciled master data. How do we start to unify all of this data using generative AIs at infrastructure accelerant, and then use that to start automating the workflow to drive higher sales volume? And that's how that kind of process looks.

Seth Earley: Yeah, I think you're on mute, Chris. That would be helpful.

Chris Featherstone: So, so when you, when you dive into it, I can only imagine that also some of the misconceptions come from the, you know, from the fud, right? The fear, uncertainty and doubt with a lot of these. Because, Seth, you brought up a good point around that this is not their core area of business technology. Their core of business, like you said, is manufacturing and industrial. So how do you take them from, you know, from understanding, like their core aspects to, you know, they know they want or need, or they have all this AI, you know, that's being articulated to them. What kind of process do you use to actually take them from? You know, from basically almost a trust, but into an area where they're like, you know, what Sahitiya? Yeah, I need what you have, you know, and I know that I need this, but I don't know how and why. So what's that process look like to get them to that point of, you know, pulling the trigger.

Sahitya Senapathy: Yeah, it's a really interesting question because, you know, we feel the strongest pull from the types of manufacturers that have actually tried to build this in house. You know, there's this legacy of companies and kind of software providers that are out there that essentially do a custom development and then do a handoff to the manufacturer. And this is kind of part of the problem in this space because you've got all these one off legacy softwares that someone has handed off to a manufacturing enterprise and they're not in the business of software maintenance. Right. This is why we had the satification of industry, because we had dedicated providers who actually maintained products. You have all these point solutions that exist. And so there's still people in the industry who are trying to build their own solutions in house because they think it gives them an advantage because they can vertically integrate the software sector. The reality is that it doesn't, it's harder. But when they kind of try building it, then they realize the difficulty of actually building and maintaining a generative AI product, let alone a traditional SaaS product. Then they come to us and they're like, we understand the problem, we know the use case. You know, you're just going to be better equipped to do this sort of thing and maintain it for us. So that's the first kind of layer there where it really accelerates the kind of process. Now the second level, how do we actually get everyone comfortable? Well, in terms of we finished discovery before going into an annual contract. Let's do a bit of a proof of concept or a pilot, right, whatever you want to call it, let's actually get this into the hands of users so that not only can the executives get confidence that hey, this is going to be a real ROI driver for us, let's actually get the people who are using it acclimated and comfortable with the idea of using this tool day to day, you know, because there's no point in putting out a generative AI tool for the sake of it if it's going to be shelfware. Let's actually get something that people are using every day and feel like are adding true value to their operations.

Chris Featherstone: Yeah, that's interesting. So, you know, part of that too is just, you know, like we do with almost use cases with these different orgs, is the crawl, walk, run, right? Figure out what this looks like from the technologist that can implement it, help them understand, you know, what that is, drive the value, work backwards from that solution, right. And then actually like you said, put up A proof of concept, some type of an MVP that makes sense so they can continue to get additional investments. So, you know, in terms of, I mean, that's pretty, pretty standard right across the board. Where have you seen like whether in industrial manufacturing where, you know, that's. Does it take a little bit longer you. In most in. In some cases or, you know, what does that look like especially? I mean, because you're dealing with these huge organizations, right? And so, yeah, I mean, I'm assuming that process is the same. But is there any nuances, you know, to this type of technology and help them over the hump of understanding?

Sahitya Senapathy: Definitely. You know, there's, there's kind of a lot of different layers to it. I think the first is in manufacturing, you're dealing with enterprises that have lots of different plants or locations that have various processes and data sources from plant to plant. And our approach is, you know, why don't we go piecemeal. Let's start with two plants to start out with and then expand this across your organization. And we've done this right. We started with two plants and now we are deploying enterprise wide. And that's where the true benefit is. But it really helps them kind of slowly move into it. And I think this is something very interesting about manufacturing because of that dynamic, that M. And a dynamic that leads to these really large companies. You know, I think, I think the second thing there is like, can we accelerate implementation? And generative AI is also a great internal tool for us as a company. Right. Can we, you know, develop proof of concepts for you faster? You know, by using generative AI to accelerate our software development process, which means that you can start seeing what the tool is like next week rather than three months or six months, which is traditional.

Chris Featherstone: Do you find that it especially manufacturers, because it's been the lifeblood of America for a long time in the US And I think it's easier, and my comment is, we'll preface a little bit of context, but it's easier for a lot of SaaS companies and technology companies to figure out what to do. And like you said, I like on your site where it says, you know, chatbots won't make you money, but endeavor will. Right. And that, you know, that has to be part of like, you know, we'll get into a little bit more in terms of handling data. But, you know, is there anything unique to maybe their, their, you know, their data? Because we have, I mean, I'm assuming schematics and we've got, like you said, all the run structure Data can be very, very complex and yet, you know, pretty simple as well. Right. And how do you, how do you help them understand, like, you know, what is this, what could and should this look like? Right. In terms of, you know, actually let's, let's, let me scrap that. Let me go back to, back to something. So Carolyn, note the time of that one. I think the one thing I want to, you know, key in.

Seth Earley: Fix it in post.

Chris Featherstone: Yeah, exactly. Fix and post. So the one thing I want to key into, you know, with this idea is, you know, we start talking about these, you know, the lifeblood of America for a lot of history and stuff is like we talked about, it's not their core to be with technology, but they're forced. Do you find that that helps accelerate or does that hinder the process at all?

Sahitya Senapathy: I think it's really interesting because to your first point, yes, manufacturing and trade, these are the core of the American economy. And one of the things that we've seen over the last several decades is we've gone from being this manufacturing powerhouse, potential Post World War II, now being more of a services economy. Right. And you know, it's interesting because we've also seen global competition rise over these sorts of things. And we're based in Silicon Valley endeavor and there's a lot of activity now around manufacturing. And you have these new age manufacturers that are trying to produce missiles or tanks, right. In the Silicon Valley style of things. Move fast, break things. But, but really the bedrock of the American economy is those traditional industries in the Midwest like the steel, the construction, the automotive. Right. And you know, they are the companies where I feel like they can most strongly benefit from AI and you'd expect kind of from the outside looking in. You know, if, you know they're based in the Midwest, you know, they're not on the coast, they're not in the West Coast, east coast, there's going to be less tech saturation. Maybe they're more aversive technology. But it's really interesting because you go into them and they recognize the problems and they understand the potential that technology has. It's kind of just a question of how are they equipped to solve those sorts of things. And coming back to that point, which is, you know, build versus buy, the folks are going to try to build it, they're going to try to build out their IT departments. They might go to individual consultants, but in the long run, it's just better off if you go to a provider who is, you know, interested in the long term maintenance of the software in actually driving the daily usage rather than, you know, doing a time and money implementation because then they're just going to drag it out. And I think it's, you know, a cultural thing. There's, there's been so many manufacturers that have just been burnt by, you know, industry 4.0 sort of, you know, you know, you know, what was the thing? Digital Twin. Digital Twin was the big thing a couple of years ago and the reality is that it hasn't especially delivered quite yet. On that ROI thing I was talking about, has it delivered top line revenue or cost savings? And you know, I think we would be hard pressed to find out, you know, if there's a single company that did that especially well. And my hypothesis, and you know, why we founded Endeavor, was because, you know, the problem isn't analytics. You know, the kind of technology behind analytics has not really fundamentally innovated, you know, in the past 10 years. But the problem is data. It's actually are we using the right data? And the question is, can we bring that unstructured data in and actually apply the kind of technology, the analytics to really deliver on this Digital Twin and the Industry 4.0 technology. And the way we, I'll give you a concrete example of this in manufacturing is there's, you know, a lot of these manufacturers again in the bedrock economy have 10, 20 years worth of data and they should be using it and you know, they've got this data that should be mined from their emails, their requests for quotes, their bill of materials that stuck in spreadsheets and you can start to do better forecasts, you can start to do better analytics, you can start to do better planning by mining this unstructured data and using it in combination with the structured data they've been using for their existing planning. And that's how I think we really get to amazing solutions for analytics for forecasting using generative AI in this manufacturing space.

Chris Featherstone: I think, you know, this is near and dear to Seth, right? In terms of knowledge management, knowledge graphs, you know, knowledge infrastructure, which I would argue in that space is probably pretty immature, right? And so, you know, I mean just in the organizations I've met, they're just barely getting to okay, we've got our data warehouse up, all fact and dimension based that we've been driving off our transactional system. And yet to your point, all that unstructured data that's sitting there and the decades and decades worth of information that's either no SQL unstructured or is structured and they aren't doing anything with it to try to find those patterns, you know, and I would argue too that maybe even not knowing the right question to ask of the data is probably more indicative of what to do with it. Right.

Seth Earley: I have a question. Give me a sense of capability. And those are good points by the way, Chris, in terms of the knowledge infrastructure and so on. But give me a sense of what a capability would look like that you leave a customer with. What are they able to do differently, especially at the end of the day.

Sahitya Senapathy: Yeah, let me kind of walk you through like a journey of what an enterprise customer might experience with endeavor. And we think about it in. Without making it a commercial.

Seth Earley: Yeah, yeah, of course.

Sahitya Senapathy: You know, we think about it in three phases, which is, you know, the first layer is data. What are you able to accomplish with data? And what you're able to do is bring in your silo data sources. You're able to start unifying data that's in different schemas, different databases, different formats in PDFs and spreadsheets in a structured format, which has historically been impossible to do without. Generative AI. The ability to tap into different formats and unify them and standardize it. Now what you can do using that data, the data is just a means is start automating that data is impossible. It's costly and it's some work.

Seth Earley: I mean we did a project for Applied Materials where we integrated 14 different data sources and gave us unified front end to that, a search based application. And it takes a lot of work, but you also have to have good content and data lifecycle management. But in any case, Gen gives you some new capabilities in that space. But continue, carry on.

Sahitya Senapathy: Sure. And you know, I think what you can do using the kind of data is start to automate. Right. And that's immediate time to value, which is you take this data now it's standardized. Can we replace something that's labor intensive for us so that our existing personnel can focus on higher value tasks? So for example, you know, you have a massive team of account executives that both interface with the customer, but they also are in charge of quoting. You know, can we shift their focus to doing more customer face time, work on order improvement, product recommendations and reduce their menial workflow for generating quotes using the generative AI technology. Now the final layer there is the analytics piece, which is really what everyone wants at the end of the day. As an executive, what you want to understand is what does my sales forecast look like? What products are going to be high in demand and what are customers going to request and now you can tap into all that data that you've automated. You've structured that unstructured data in the PDFs and email and spreadsheets. Now let's start driving insights and analytics so that executives can make better decisions. And that's kind of the three step phase that we think about.

Seth Earley: Okay, and so again, when you come back to what's going to be different, can you give a specific example? I mean, in terms of you mentioned sales forecasting. So you're saying that it would be easier to do sales forecasting with fewer steps. What else would, would there be an impact on the customer experience? Would there be an impact on the employee experience in terms of, you know, again, integrating these different structured and unstructured data sources? So, you know, is there a service component of this? Is there a self service component? But what are the kinds of things that you're seeing? You can really get traction with customers.

Sahitya Senapathy: Yeah. So at the automation layer, again, it's immediate time to value in the sense that for quoting, right. You can immediately push out these quotes, which means traditionally your customers might have to wait a week or, you know, four days before they get a quote and they're kind of just twiddling their thumbs trying to get, you know, what is this going to cost me? Now you can start to immediately respond and almost get a frictionless, touchless experience that your customers are really going to appreciate because it's faster. Same for your inside sales reps. Right. They're going to be able to interface with the customer and actually get more face time so that your customers are talking to people instead of talking to a machine. Now the analytics piece is a bit different because it's not only the ability to forecast better, but it's the ability to amalgamate all this information you hadn't been able to do before because like you said, you can do it. But it'd be very costly to start using that unstructured data. And now you can do so in the same sense of traditional analytics where you can do big data analytics, but you can do so with unstructured data in the same way that you'd be mining numeric data now with the non numeric.

Chris Featherstone: Yeah, because that time series data is going to be super critical for planning and for, you know, predictive, you know, predictive planning. But you know, the supplier piece. I think what's interesting too, Seth, you know, just in terms of, of a lot of this is in Sahitia centered around, I feel like especially in a lot of European, a lot of supply Chain. A lot of these types of technologies, these classic legacy technologies, they over index on planning, they over index on potentially logistics. Right. And things like that, like supplier planning, you know, you know, goods on hand, all that kind of stuff, but they don't on the actual real time manufacturing piece. Like, you know, what does that look like in terms of, of the flow, what's happening, you know, real time, what's going on right now? And then how does that give me the ability to predict what my output is going to be at the same time, look at it from the perspective of supplier management coming in, looking at goods on hand from their perspective and what that can give me. Because I think we get like the planning right, but it's all those other aspects and the connective tissue between them that lose focus. Is that a fair assessment or is that, you know, like, am I missing it?

Sahitya Senapathy: No, I think that's absolutely right. You know, I think planning is the end all, be all. Everyone wants to do better planning. It's just a question of do we have the right data. Yeah, and it's, it's about bringing those different processes, those siloed data sources and connecting the different ERP systems really hard because of the lack of standardization, using the unstructured data on hand to drive better decisions.

Chris Featherstone: Yeah, because you have all that real time stuff hitting the conveyors and coming through in a, you know, a classic, let's say, you know, supplier type perspective. But then, you know, when that gets into the actual manufacturing floor and you know, and then you like, you see your com bonds decreasing and then that, you know, adding to the, you know, the output of the products that you're, that you're doing part of that's, you know, I mean you, you may be the best part about this is, you know, your suppliers are feeding you and you may be a supplier for something else. And part of that too is making sure that predictability for your downstream is also there. And then Seth, to your point, you got all of the HR and employees and engineers and everybody that's centered around the gravitational pull of what the core product is to actually generate all this stuff and then get answers quickly because it's. The most critical thing is, hey, what in the hell is going on within my, you know, company right now? And I have no idea, you know, so I need to be able to see that. But anyway, yeah, it's interesting.

Seth Earley: Yeah. And, and you know, when you think about what you're trying to do there, you're trying to speed up the information flows throughout the enterprise and get people the right information that they need when they need it. The old mantra of knowledge management and personalization and contextualization. How do you, how do you go about fixing that data? You know, you talked about the fact that it needs to be standardized. What does that process look like?

Sahitya Senapathy: So we do what's called a forward deployed motion, which means we embed very deeply with the client. And often we go in person to visit our enterprise clients. And I say we don't do a deal without actually visiting our clients. And it's very important to actually go meet the folks that are in charge of the data, to view the data, to actually kind of like know, start, you know, experimenting with it and then look at the schemas, look at the columns, understand the process, and then start unifying it into one thing that is standard. And then you can really accelerate the process of this schema standardization, this data cleaning using generative AI, which I think is, is new, right? It's laborious to do without. But, you know, you have a long history of these ERP companies that focus on standardization. But if, if you can now start to actually have a semantic understanding of a process and of a data table and the associate schema, well, then now you can start to reconcile different tables. When there are variations from plant to plant and location to location, business divisions.

Chris Featherstone: It's interesting. So do you see that across, like, you know, the, you know, more relevant across, let's say the sales quoting side of things or material planning or, you know, just pricing or entry? Where do you see that as the most applicable?

Sahitya Senapathy: I. We really see that sort of thing on the operation side for manufacturing, you know, there are going to be variations on the sales side for sure. You know, there's going to be different product catalogs, there might be different processes, but for the most part, ideally you're, you're quoting the same way from plant to plant at least 80, 90%. But when you get to operations, it's kind of the wild west, because again, these companies have probably been around for a long time. They're almost like you got four or five different companies under the same umbrella and they've been doing things for the same way for the last 30 years and they're still operating that same way. And when you look at a different region, you might look at the division in the west coast and the east coast, they may have completely divergent strategies for producing a particular good, which means, you know, figuring out the materials that you need for it, figuring out how to actually use those materials to produce good, then Scheduling them on your shop floor and the idiosyncratic processes that exist from plant to plant and location to location is really where the kind of fractured systems come into play. And the lack of information,

Chris Featherstone: I would. Argue, how many of those are just managed by a single spreadsheet in each one of them. Right. With their own process, their own models, whatever, all done in a, in a spreadsheet. And with the worst file system ever. Email. Yeah, right.

Sahitya Senapathy: Yep. I, I would probably guess 100%, you know, I would be hard pressed to find a single manufacturer that, that does not use Excel, which you have these.

Chris Featherstone: Huge systems like SAP and Oracle and yet everything's still sitting in Excel as their day to day work productivity tool.

Sahitya Senapathy: Right. And that's what they're going through, which is, you know, is scary at the same time as that's just what, you know, how things have been done. So continue to do that.

Chris Featherstone: Yeah. Well, it's funny because you could probably just

Sahitya Senapathy: ask someone, you know, in Silicon Valley here, hey, what do you think is the most used software in manufacturing? And they might be like, hey, is it Oracle, is it SAP? But the reality is, you know, you know, I'm from Texas, you know, we've actually been out there in the real world. It's like, it's Excel, it's actually Microsoft Excel and Outlook.

Chris Featherstone: And then all reports are done through PowerPoint.

Sahitya Senapathy: Yes, exactly. Yeah. That's interesting.

Chris Featherstone: So, you know, I don't know if it's a, it's, it's a good time to get into it because I think one of the other aspects to, you know, Seth, that we often, you know, kind of dig into with these is, you know, do you see? Or, or maybe the better question is with a lot of these organizations that are also still trip, you know, that have trepidation about AI, what do you see about them in terms of insecurity around data sharing and data privacy and all those kind of things, because I'm. Yeah, they kind of go hand in hand. Right,

Sahitya Senapathy: that's, that's a really great question because it always comes up on that first conversation. In fact, I think it's not even about cost, it's about data. You know, like what's going to happen with our data? Because you have all these horror stories. For example, like Samsung, you know, employee puts in some proprietary data into ChatGPT and there's leakage there. Right. It's not protected. And you know, the first thing we get is like, you know, who owns the data? And our answer is, you own the data. Right. We're not in the, we're not a Facebook, you know, we're not in the business of reselling or sharing other people's data. You own the data that goes into the model and we don't do any data sharing. And you know, and usually that's, that's the biggest concern people have are our insights, the learnings from our data going to improve my neighbor's business. Because we don't want that and we don't do any of that. Even though, you know, there are other companies that are in the business of doing that, you know, doing a group pool of data and improving their models based on that. But, you know, our systems are architected such that you don't need to. The second layer of that is your traditional, you know, security and data management best practices like, you know, do SOC 2 type 2, you know, make sure that you're hosting everything through, you know, a good cloud provider. Make sure that you are hosting things in the right data centers and doing the right kind of encryption at rest and in transit and hosting data in the right kind of sources. And without those sorts of best practices that we have a great lineage for, it doesn't really make sense to be in the business of enterprise sales.

Chris Featherstone: So define irony that we're worried about our data and data security. However, we haven't touched that data in 20 years.

Sahitya Senapathy: Right. And this. Yeah, go ahead, go ahead. No going. You're good, you're good.

Chris Featherstone: It's interesting because this also leads to the next question of prem. Right. Which is everyone want. Yeah.

Chris Featherstone: So. Yeah. Because now it's not only is it, like you said, not only is it, we haven't touched it, but we're super, we're super insecure about it because we don't want anybody to touch it. But it's sitting in, in a bunch of prem systems. Going back to your original point of who the hell is going to be there to manage all of that and keep it updated. And it's still on, you know, disc. Hella maybe it's on, you know, magnetic tape somewhere in, in Iron Mountain. You know what I mean? And they don't know why they do it, but it was the best practice of the day. Right?

Sahitya Senapathy: Yeah. You know, when I think about it, there, there are definitely some areas where it makes a whole lot of sense to have fun. Right. If you're, if you're, you know, one of the big auto companies and you've got some proprietary designs and you really don't want anyone leaking those designs. Yeah. Like, like keep that prime if you're Coca Cola and you've got your secret recipe, obviously it makes sense to lock that up. But if we're talking about, you know, ordinary data, you know, having it on PREM and then trying to deploy a local model like a llama or a Mistral, and bring that in from a PREM standpoint because we, you want generative AI, but you still want to maintain your data sovereignty in your, your own hardware. It's just extremely challenging. And you know, the kind of compromise that we often get to is, hey, let's deploy this in your VPC in your virtual private cloud, right? It's a hybrid PREM solution. You can deploy a language model in their cloud, but they can maintain the kind of network and security that's associated with it. And that's a happy medium for these large enterprises that care about maintaining the kind of data privacy and sovereignty.

Seth Earley: Yeah, yeah. I mean, at the same time, you know, there's so much that's in the cloud that people are kind of used to the fact that you have to secure your, your data and you're using cloud providers, you're using, you know, software as a service and this is in a way no different. Right. Because you still have to, you still have to have compliance with security standards. You had a very interesting set of experiences. Sorry about the cat in the background. I don't know if you can hear that, but no, you can't. It's amazing. It's really lo good microphone. So you had some interesting experiences in terms of internships and early in your, in your education. You want to talk a little bit about that? What did, what did you kind of go through when you were a teenager and you did some interesting internships. Let's just reflect on that a little bit. You want to tell that story?

Sahitya Senapathy: Yeah, happy to. I started coding when I was 11 and nowadays this is not entirely uncommon. But you know, unlike, you know, a lot of my friends who got started early and you know, built a video game or a mobile app, the first app I ever built was for fema, the government agency. And let me, I can tell you the story here. You know, we had a tornado rip through Texas from Dallas and you know, a lot of tornadoes come through and the roof of our house got ripped off and 11 year old me is, you know, talking to my parents like, hey, why haven't they fixed the house yet? You know, what's going on? And they're like, oh, you know, like fema, whatever, X, Y and Z. So you know, I took my parents computer and I emailed the guy at fema. And I was like, hey, you guys got to fix our house. And I had my parents drive me, you know, 40, 50 minutes down from Dallas to Denton. And I went and pitched the guy at fema and I think there's like some news article about this, but I. Was like, you actually, what did you pitch to him?

Sahitya Senapathy: I pitched him. I was like, you know, why is it so slow? And the guy's like, you know, all of our processes are running on paper. You know, we have a lot of phone and communication overhead. And I'm like, you guys should digitize this. And so I built them. You know, the first thing I ever built was this.

Seth Earley: So you walked in the door as an 11 year old. Like usually you, usually you hear, you know, they didn't know how old they was. We never did anything on video. They just saw email. But you actually walked in there and they see a little kid or your age.

Sahitya Senapathy: There's a picture if you search my name up with me at, I don't know, some like four feet tall and this, this actual adult fema. Cheap. And no, it's, it's a crazy picture. And I go in there and, you know, it's, you know, one of the most incredible starts of the journey there because that's how I got started coding. You know, that's why I'm interested in traditional industry. I don't know if you've got that picture pulled up, but you can see.

Seth Earley: If you want to pull it up, it would be great. I will it. I'd have to spell out your name here. Oh, here it is. Let me just. Yeah, I

Chris Featherstone: got it. You got it. Sixth grader helps develop a plan for first responders. Yep. Yep, that's the one. Yep. That's great.

Sahitya Senapathy: You know, we did

Chris Featherstone: not unlike, you know, other government agencies and stuff, I was pulled into pandemics, which again, you know, in 20, I think that was 2007 or eight, all a paper process like you said. And it was amazing to me that that's what was acceptable. How do you part and parcel to this is how are you able to respond real time to needs of people which arguably now we're into the emergency aspects with using paper and phone tag, that is processes that could have started in the early 1900s. This what's acceptable. Right. And I would, I would argue probably still have some of these processes in place today, but anyway.

Sahitya Senapathy: Oh, definitely. Especially in those, you know, DoD and defense realm, even in manufacturing.

Seth Earley: So you had a team, you had a team of six sixth graders it. Looks like,

Sahitya Senapathy: yeah, me. Me and a couple friends just built this application, and it was great. And we actually won, you know, $5,000 grant from the Army. And I never got an allowance as a kid, so that was the most money I'd ever seen in my life.

Seth Earley: And so then what did you do? What did you do next? You said you got this something for the army.

Sahitya Senapathy: So I got this $5,000 grant from the army and, you know, educational stipend, and I got pretty into math, and, you know, I started Math Olympiad prep, and I became a Math Olympiad semifinalist when I was in middle school around this time. And, you know, naturally, you know, that lends itself to military. And when I was 16, I joined the Air Force Research Lab at Kirtland Air Force Base, where I worked on a team of PhDs doing deep reinforcement learning. And I put out a paper as well that's out there on, you know, how do you apply. And deep reinforcement learning at the time was all the rage. You know, AlphaGo, right? You know, you have. You had autonomously working chess and beating Lisa Dong, who's the. Who's the go, Grandmaster. And, you know, of course, the military was interested in, like, how do we apply this to our autonomous drones, right? How do we actually have, you know, cool tracking software? And I got to work on that, which is awesome because I grew up watching Star wars, and I'm a prequels guy, by the way, not a sequels guy. Prequels get too much hate. I love Revenge of the Sith, but, you know, I'm working at this Air Force base, and I worked my way up to having a sequel of a clearance, and I got to work on what I would watch movies about, and that's how I really got into ML from software.

Seth Earley: Neat. That's great. And then what did you do after the. You did stuff with the Air Force. And then what did you do after that?

Sahitya Senapathy: So. So decided to actually go to college after spending a lot of my high school just around a bunch of grownups, and I. I went to Penn. And, you know, I. What I like to say is my dad's a businessman, my mom's a scientist. Wanted to make both of them happy, so I decided to do the Wharton degree and the computer science degree so that neither one of them got too upset at me, even though I think they both wish I was a doctor still to this day. While I was at Penn, I interned at aws, and I got to work on Amazon Q, actually, and that's where I got to work on LLMs and you know, we were working on this before Chat GPT even came out and was working on LLM infrastructure. Was. Was building some of the core infra in rust for semantic search, which is really interesting because I'm a Python guy, you know, I've done ML research at the Air Force and you know, interpreted languages. And then you go to the low level of rust and it's like a completely different ballgame. But decided I didn't want to be a core software engineer after that. So went to Palantir and that's where I got into the manufacturing space. You know, at Palantir, I worked with a Fortune 200 manufacturer, you know, did one of the first generative AI implementations there was. Was going around the country to different plants. And that's where I really fell in love with the space, you know, because there's so much in the manufacturing industry. We talked about it. It's core the American gdp. It's the real world. And that's what kind of led me to starting endeavor.

Seth Earley: What kind of manufacturer were you working with?

Sahitya Senapathy: Yeah, I don't know. So I wasn't working with the manufacturer for aws, but I was at Palantir, but I don't know if I'm allowed to say, just a field. Like what. What did they ballpark? Like what kinds. Was it diversified or. It was. I think it was. Is a process manufacturer.

Seth Earley: Oh, okay. Yeah, yeah. Chemical plant or something like that.

Sahitya Senapathy: Yeah, that type of thing, you know, Pharmacy. Yeah, very. And it's smelled when you go to the plant, the smell just. Yeah, it just sinks in.

Seth Earley: Right, right, right, right. So what do you do for fun? What else outside of work?

Sahitya Senapathy: Good question. I recently picked up water sports. And you know, it's funny because I was. Even though I grew up in Texas, I was born in Michigan, so you'd expect me to be a big skier, but I was. I never really got into it, but I picked up water skiing and I was telling Chris this the other weekend, but, you know, we were. I was water skiing and I'd never done it before, so I kept falling over and the skis kept falling off. But eventually, you know, I think I got it and the boat's taking off and I'm right behind it holding on, and the guys hold on and like hold your arms straight, don't let go. And the skis come off and I'm like, okay, well, don't let go and hold my arm straight. And I'm just flailing behind the boat like it's dragging me around the lake.

Seth Earley: Oh, no, yeah.

Sahitya Senapathy: No, I mean, incredible experience, though, because eventually, eighth try, I got it. And it's. It was unlike anything I've ever done before. You know, I've done. I've gone skydiving before, which is awesome, but, you know, water skiing, it's like, in the moment, you know, and, you know, I'll be doing a whole lot more of that, and maybe I'll get. Get into skiing as well.

Seth Earley: That's great.

Chris Featherstone: Yeah, we're gonna. We're gonna try to, you know, get him on to the other water sport, which is snow skiing, you know?

Seth Earley: Yeah. Frozen water sport. Yeah. Well, this has been great. I've really appreciated your time, and it's been a lot of fun chatting with you. Thank you for participating and joining us on today's podcast.

Sahitya Senapathy: Thanks for having me. It's a pleasure and love talking about what I do. You know,

Chris Featherstone: your energy is palpable, my friend, and it's. It carries through. So, yeah, it's so fun to hear the excitement in your voice around just. All the

Seth Earley: thanks to folks tuning in to today's podcast. People can find you on. On LinkedIn and then endeavor. AI.com is the name of. Is the URL of the company. E n-d e a v o r a I.com and thank you again. Sa. It's great to have you.

Sahitya Senapathy: Thanks, Seth. Thanks, Chris. Thanks. Thanks, everyone.

Seth Earley: All right, cool. Okay, bye now.

Sahitya Senapathy: Cool.