Despite how deeply we’ve traveled into the digital era, employees at enterprise companies still find themselves devoting hours of their day, if not days of their week, poring over text documents looking for data and answers to questions.
AI and machine learning technologies make automated solutions possible, but difficult to implement, often requiring a team of developers and data scientists.
Ryan Welsh aims to make the solution much simpler with his company, Kyndi, which offers an AI-powered reading platform API, allowing companies to quickly and easily build applications to analyze documents and retrieve information.
More information: https://kyndi.com/
Ryan Welsh is founder and CEO of Kyndi, a global provider of the Kyndi Platform for the Natural-Language-Enabled Enterprise, an AI-powered platform that empowers people to do their most meaningful work.
Kyndi was recently named a Technology Pioneer by the World Economic Forum for its Natural Language Search, which delivers highly relevant and context-driven information with unprecedented speed, keeping humans at the heart of critical problem solving. For more information, visit https://kyndi.com/ or follow @kynditechshow more
Kyndi is a global provider of the Kyndi Platform for the Natural-Language-Enabled Enterprise. Kyndi’s Natural Language Search, a major component of the platform, delivers highly relevant and context-driven information with unprecedented speed, keeping humans at the heart of critical problem solving.
Backed by leading venture investors including Intel Capital, UL Ventures, and PivotNorth, Kyndi was recently named to the Forbes Midas List for visionary investments and was named a Technology Pioneer by the World Economic Forum. Customers include Fortune 500 pharmaceutical companies and government agencies. For more information, visit https://kyndi.com/ or follow @kynditech.show less
DISCLAIMER: Below is an AI generated transcript. There could be a few typos but it should be at least 90% accurate. Watch video or listen to the podcast for the full experience!
Ryan Welsh 0:00
Well, then you now find the answer to your question, you can now go go solve the problem that you’re trying to solve for able to comb through 1000s of documents and extract all the relevant information that you’re trying to extract, you now have that information, and you can go Go do your problem. So it’s really trying to just remove everyone away from doing the mundane, low value work that can drive you crazy, to really just focusing on doing your most meaningful work.
Alexander Ferguson 0:30
Welcome to UpTech Report. This is our applied tech series UpTech Report is sponsored by TeraLeap. Learn how to leverage the power of video at teraleap.io. Today, I’m joined by my guest, Ryan Welsh, who is based in San Mateo, California. He’s the founder and CEO at Kyndi. Welcome, Ryan, good to have you on. Thank you very much. Thanks for having me. Now, Kyndi is a natural language technology platform focused on serving enterprise. This includes industries of manufacturing, Life Sciences, insurance companies, government, and you’re really working with CIOs, CTOs head of research, helping them implement natural language technology into their enterprise to help them understand what the heck is going on, and be able to search and get information faster. And we’ll dig into what all that means in a second. I’d love to kind of hear from you, Ryan, though, help us understand lay the play the help us understand the playing field here. What’s the challenge that you see in these industries that you’re helping serve? What are they facing right now?
Ryan Welsh 1:31
Yeah, it’s, there’s, there’s actually two layers to the to the problem. The first is from a user perspective, so 77% of enterprises today still process text data manually. Like that’s a, that’s a crazy thing to, to think about it in our, in our modern digital world,
Alexander Ferguson 1:50
when you say process text data manually, like you just just help me give me an analogy or another process text
Ryan Welsh 1:56
data, look through all these documents and pull out all the interest rates, right? If it’s a financial document, like literally 2.5%, you know, X percent plus library or something like that, like, pull out all these these interests, like, Okay, so now they’re doing all that, that manually. Look, through all this text data and find answers to questions, you know, stuff like that. I mean, I mean, they’re spending so much time and effort looking through all this, this text data, it’s because machines can understand natural language. And so there’s the end user problem, that is, employees spending anywhere from one to two days a week looking for answers in an extracting information from text data. But then there’s a problem below that were actually getting machine learning to work in the enterprise is incredibly hard. And it’s hard because you need to hire a bunch of AI talent, then there’s not many people in the world that know how to use these algorithms for specific problems. And it’s also difficult because you, it requires a lot of labeled training data, which enterprises don’t have. And so there’s there’s the end user problem. And then there’s actually the data science, the it the platform problem. And so, you know, as a platform company, we like to solve the platform company. So that are the platform problems so that people can then use these technologies and embed them in applications to solve end user problems.
Alexander Ferguson 3:25
Yeah, highlight the, the underlying the underlying issue of Okay, we are spending a lot of money, energy, people resource time, on entering all this data manually. technology exists to solve it. And it’s not terribly new, it’s been around for a couple of years, but actually implementing it has been the challenge. And that’s where you guys are trying to address it. Let’s hear a bit more about your your story, though, of like, seven years ago, right, that you started Kyndi?
Ryan Welsh 3:49
That’s right. That’s right. So seven years ago, I started I had an experience prior to starting the company, where we had to read through a lot of financial contracts. So I was working for a law firm back back in New York, and as a securities law firm, and during the financial crisis, you know, when all this stuff went into bankruptcy, we had to reread all the, the contracts to understand what the heck was actually in them, and ultimately try to unwind these things. So I mean, quite literally, I would spend years like there was a three year period where I just was just, you know, combing through, through documents, read it, reading it, I mean, it’s
Alexander Ferguson 4:27
like, I hated doing that. I
Ryan Welsh 4:29
imagine you hated it. It was you know, I, I had a graduate degree in quantitative finance and mathematical economics. And here I am spending years just reading through financial contracts. And when you think about it, it’s it’s you wonder, you know, why don’t we have technologies that can read and understand natural language. And then when you think about it also is, as humans, we haven’t been able to increase our reading speed. We’re essentially reading at the same rate that our parents and grandparents are reading at. And so effectively, we’re the bottleneck in them. During production process, right, we can store a lot of data, we can move it around super quick. But as soon as it hits our eyes, we’re still reading at the speed that people read out read at in the early 1900s. And so,
Alexander Ferguson 5:14
so this is bad, you’re experiencing yourself back in like 2008 2009 2010. And you’re like, I’ve got to be a better way. But there’s still a few years between that and starting Kyndi. So what did that Yes
Ryan Welsh 5:24
Is it then there’s there’s business school from 1111 to 13, which is, which is which is always always good. So I went back to the numbers nerd and did get my MBA, I got out and I started working for a consulting firm in in in Washington, DC that was run by senior intelligence agency officials. And they are always helping commercializing technology coming out of national laboratories. So very advanced technology, low orbit, satellite, quantum cryptography, cyber capabilities, just all types of unique stuff that we develop in our national labs. And then in 2014, I had this thesis that I believe that by 2025, all knowledge workers would need some form of AI powered workflow, and specifically around natural language technology, and helping us sift through this information so that we can get through it faster, and hopefully spend all of our time doing our most meaningful work. So I started Kyndi in 2014,
Alexander Ferguson 6:25
to create a thesis out of your business school, and then you’re like, let’s turn this into a business.
Ryan Welsh 6:29
Yeah, exactly, exactly. I actually I actually met a gentleman who was an angel investor in the DC area. And I kind of had this idea of this, this major thesis. And what’s funny was, he said, Ryan, that’s interesting, because I had met a scientist, that’s looking for a someone that knows math really well and knows business. And so this gentleman actually set me up with my technical co founder on a founders date. And we met and we hit it off and started Kyndi, you know, week or two later.
Alexander Ferguson 7:03
That’s a brilliant strategy or tactic of finding your founder just sharing a thesis with others. And then they can say, I know someone else that has a similar interest. And that creates that that bond for a co founder.
Ryan Welsh 7:17
Wow, exactly. And I was I was very lucky to meet his name. His name’s Arun Majumdar was very lucky to meet him. Because he had been in the AI space for for four years. I mean, for 30 years, he had been mentored by some of the pioneers in the field, including Marvin Minsky, john McCarthy down at Stanford, he was a part of the party, McCarthy’s former reasoning group down in Stanford, mentored by people like john Sela. And so the reason why I bring that up is I was able to get a crash course in AI, from, you know, people that really knew the space. And it wasn’t like, hey, this, these machine learning techniques are now starting to be productive, let’s apply them to everything. It was people who knew the entire, you know, history of artificial intelligence, really bringing the best capabilities to bear.
Alexander Ferguson 8:09
So you’re bringing the use case, they’re bringing the technology and the insight together to then address this problem, when you shared this with your co founder of this idea of applying in this particular way was there like yes, this is a great idea or to take a little bit time to, to come to a fruition on again, was,
Ryan Welsh 8:29
was was interesting. And I think I’ve been pretty good at this, in my career is having big macro theses. And so and so when you come through, and there’s these, you know, artificial intelligence researchers that are really in the weeds on on doing research, and you come through and you say, Hey, I think now is the time to transform the world. They kind of like look up and go, Oh, yeah, that’s why we’re doing this. You know, and because they’re so in the weeds, and so when you when you kind of paint the picture of, hey, now’s the time to, to leave the lab transition it to applying it to specific problems, to get really excited. excited by that. And of course, I get excited by the technology as well.
Alexander Ferguson 9:12
What’s the What year did this happen? Was this in 2014? That Yeah, yeah. Early 2014. Okay, and then and then from there, you guys hit it off? And what happened next, and it just off to the races? No,
Ryan Welsh 9:25
no. So it’s, it’s a lot of it’s a lot of research. And so what’s interesting about the natural language, technology space, is, you know, when when you see AI start to take off in 2012 2013, after image net. It’s all around computer vision. And one of the big knocks on deep learning was that it wasn’t good at understanding language. And language is a really difficult thing to understand. And it wasn’t until you know, 2018 where with the release of Bert that used To see deep learning actually starting to be applied for real meaningful language problems. And so I bring this up, because for a good period of time, it was research. It was, you know, how do we apply these specific technologies to understand the natural language, and then also matching them to business problems? Because a lot of times, I think what people really don’t understand with with AI is, you know, it was developed in academic institutions, a lot of them. And it wasn’t thought about or at the time, you’re not thinking about how does this then transition to a business case? And is it applicable to to a business case? So you know, take take, for example, deep learning, one of the biggest problems with with deep learning is that it requires an insane amount of data to train on labeled training data. Well, guess who doesn’t have that? businesses? Right, especially when you require web scale data to train on? And so you know, and so people didn’t think about, alright, well, we need to build systems that learn on less data. And they also didn’t think about all right, well, we need to build systems that are explainable. And so when we started navigating the problems of the technology, we actually had to develop new techniques to make it work within the the space that we were going after, which was within the enterprise.
Alexander Ferguson 11:30
You mentioned a bit ago, Burks, which if I google quickly that Google a created that, is that correct? Yeah. Was it 20? What year did they ship in
Ryan Welsh 11:41
late 2017, the paper came out. And it really kind of hit the press, probably early planting.
Alexander Ferguson 11:46
And it’s an overall I guess, algorithm are a way that one could could then understand natural language.
Ryan Welsh 11:53
Yeah, yeah, it was, it was a really great breakthrough. Where they started to understand the context of of language more. So they looked at both the right and left side of words to really understand what does this word means? So as an example, the classic example is bank. Is it a river bank? Or is it a financial bank? Right? It’s it’s hard to understand unless you look to the left and right over the words to understand the context of those words. And there was also some parallelisation, that it enabled, which allowed us to train on a lot, much larger data sets and get that context and really start to understand language model,
Alexander Ferguson 12:31
bi directional encoder representations from Transformers for those Yeah. But basically, you paint this beautiful picture, the academic side, they create these algorithms, and it’s beautiful. It’s nice. But Alright, now how does it cross the chasm and make it to where the rubber hits the road, and people can actually use it in a business use case is this where then you’re looking around, you’re seeing these algorithms, you’re seeing these existing solutions, and then building on top of it, your your additional
Ryan Welsh 13:01
AI would say, that’s what that’s what’s interesting about Kyndi is, is given the the brain power that we had early on with with the researchers, we started pioneering new new approaches, we’re starting to solve long standing problems in the AI space. And so you know, one of the one of the interesting things about AI, and I’ll make it super easy, and AI experts, you know, don’t criticize me a lot. But it’s super easy is there’s kind of two two main fields. There’s the connectionist approach, the machine learning the deep learning approaches, and then there’s these symbolic approaches. And it turns out that the merits of each overcome the demerits of each field. And so if you can actually fuse the two pack, fuse the two paradigms together, you have a really rich and robust system. And what’s interesting about AI research is that what I found and this is an outsider’s view is that the the researchers in each camp net don’t necessarily like each other. And so you either work in one or you work in the other and you believe one is the path the jet artificial general intelligence and you you’re not you’re not the other and and so when we came, came through and said, Hey, we really don’t care. We we want to create a better system. We really thought about how do we fuse these these two paradigms together? And how we how do we you know, really push the state of the art and that’s really what we achieved here at Kyndi. And recently, were recognized as a pioneer by the World Economic Forum for pioneering this field of AI that’s now being called neuro symbolic.
Alexander Ferguson 14:46
neuro symbolic who just combines the two pieces together how to understand that the timeline though, so you know, 2014 you guys get together you start the research, How many years did it take them to start to make those connections and build this together?
Ryan Welsh 14:59
Alright, For years to prove everything out. So you have some initial, you know, ideas about how things work together, you start to prototype them spaghetti code, so not not not production, you’re just hoping hope, hoping it works, your ideas works, you start to then prove it out on on on business problems, we were working with the US government very closely on some big problems that they had an understanding, you know, text data, and you start to just prove out that your ideas, you know, could work, or then start to work in the wild. And that happened for for four years. And then after four years, I mean, this is early 2018, you know, we bring in a great CTO that can actually transition science to an enterprise grade platform that enterprises can can use, then. So we’re now around 2018, where we really started to
Alexander Ferguson 16:00
build a product out of the underlying algorithms, do you really did spend four heavy years on the research side, still bringing it to a final product? versus just taking something existing and then turning into a product, you’re actually truly building this this new product? But but then you need to still make that transition to Alright, we’re How do I sign up? How do I actually just apply it to my existing business? Did you immediately know the industries that you were going to be focusing on serving?
Ryan Welsh 16:29
Yes, so so so we did, we did do a lot of customer discovery research around which which industries to go after. So you have these kind of natural hypotheses of this industry as a bunch of text data, you know, this industry doesn’t. So let’s go over here. And what we found was was one, we narrowed in on industries that had a lot of text data, but we also narrowed in on industries that had already started to see success with data science platforms. And so there’s been a lot of success in applying machine learning, what I’ll call advanced statistics to structured quantitative data. How many red shoes Am I going to sell in Kentucky next quarter, right? Like you can collect a bunch of data on that. And you can predict and you can be, you know, good at predicting in a statistical way, then optimize it optimizing certain supply changes in the inventory, inventory, and all of that, just just around that. And so there was these these industries that were seeing success in the lower hanging fruit, that we started to say, they’re now climbing up the tree to get more and more value out of the the harder problems, then we found that the industries that had already realized success with machine learning on their quantitative data, we’re now starting to apply technologies to the harder problems like computer vision and natural language.
Alexander Ferguson 17:56
I’m curious for you during that four years, and then as you got through that, how did you feel? I mean, you’re not at you’re not actually in the weeds coding anything. We’re creating it. So where were you feeling? Oh, yeah, this is coming along swimmingly? Like, I’m just curious as, as an individual, as the leader there. What was that like for you?
Ryan Welsh 18:15
Well, there’s specific outcomes that you’re that you’re driving towards. And so you know, we still generated revenue during that that period. So we’re still selling to customers. So it’s not just like pure, like research at an academic institution, like I, you know, it was it was a business, we went out, and we sold a product products to customers that were willing to take on very early research to prove it out. Because if if it worked, the upside was was massive. And so the bet for them was asymmetric. Right? Like, alright, I’ll give you $150,000 for, you know, this software for a year, and it’s full of, you know, just academic code. But I’ll realize $30 million in value, like the best asymmetric,
Alexander Ferguson 19:07
yeah, is if anything, they’re your beta clients, whoever you want to call it, just being able to test that they get the up the huge upside. It’s an interesting recipe for others. Also, entrepreneurs are looking to develop new technology companies of like finding those are willing to be your pioneers. Sure, early stage or early interest, was it difficult to find them or those who are interested in playing with it? Or do you see a ready desire for implementing this type of technology?
Ryan Welsh 19:34
Yeah, well, there are certain industries that are willing to make those bets. And there’s there’s the US government is is a great one. So people say hey, it’s really hard working with the US government. There is a lot of red tape I you know, that there is but they’re willing to take these bets and and they’re willing to fund things too. Prove it out. Because I mean, if you look back at the history of research and innovation in the United States, it actually comes out of the God, DARPA, stuff like that it’s only been in the last 20 to 30 years that private institutions have really started to fund all that research. And the government no longer does does that research. So they still have a long history of funding very innovative things, and they’re willing to take those bets. And there’s also companies in the private or in the commercial sector, that are also willing to take those bets, but you need you need to find them. And they are, they are hard to find, you got to go out and embarrass yourself a little bit and, you know, pitch a lot of people and you know, you’ll find someone and they’ll pay for it. Yeah,
Alexander Ferguson 20:43
I’m curious, like how you’re processing? Do you already have those connections? It was just natural for you? Or were you having to reach out beyond your network or specific people that helped you make those connections?
Ryan Welsh 20:54
Yeah, what was easy reach out beyond your network, you need like a cast of very, very wide net, to find a few crazy people that are that are willing to, to take this on. But but there’s, there’s, you know, when you stay to an idea of a future state, there’s, there’s people that that like, latch on to that. And so, you know, when you say, you know, one of the things that I kept saying was, Hey, you know, we’re the bottleneck in the modern production process, like we can’t read understand information any faster than our parents or grandparents. And I believe that there’s technologies that will just significantly increase our capacity to understand that information. Some people that were like, Whoa, I get it. And then they’re like, I get it, like, I don’t actually get it. But I kind of get it, and they’re willing to take that bet, because they see the transformation. You know, they they see how, you know, you could go from employee spending one or two days a week to employee spending, you know, five minutes, you know, and that’s just that’s just huge across a 200 or 300,000 person manufacturing company. And so the the, the gains can be can be massive, you just got to really find your believers and the people that are that are willing to believe in the future that you’re trying to create, then, you know, work your butt off to, you know, help make them look great. And realize that future
Alexander Ferguson 22:23
makes me think of the adoption curve. Where it sounds like you’re finding those early adopters, those really innovators, those who love and see the value in the future. That’s helpful when when you’re when you’re launching and be able to show the potential of it. I’m curious, are we getting to the early majority, and are we able to cross this chasm? Or is it a far along way off? Still?
Ryan Welsh 22:44
I I think it’s still it’s still a long way off. I you know, forgetting the eggs, the exact categories, but but, you know, I think there’s one even before early adopters like what’s what’s the I forget the very small one before you get to the early adopters is before the chasm. But there’s one even before
Alexander Ferguson 23:02
innovators, early adopters, early majority,
Ryan Welsh 23:06
candidly, I think we’re I think we’re still at innovators.
Alexander Ferguson 23:09
Okay. Not even early adopters yet,
Ryan Welsh 23:11
where I don’t even think we’re at we’re at early adopters yet. Yeah, maybe I mean, maybe just in the last quarter to, like we’re starting to get into, you know, early adopters. But I honestly think we’re pretty far off from from,
Alexander Ferguson 23:26
what’s the robot? Like, what’s, what’s stopping the early adopters? So we say, yes, let’s jump on this.
Ryan Welsh 23:33
Yes, it’s a tick search as as an example. With AI, you can create a search capability that where users can ask natural language questions, so type in questions, as if they were asking a human, then the system can return an answer, or in or in some systems return a snippet of text that contains the answer. Most people will say, Hey, I have I have Elasticsearch. And you go, like, there’s a huge difference between returning links to documents and returning the answer. And, and, and, and so and so with with, I think what a lot of people don’t think about is, is what is the amount of time that my people are spending on when they search for documents, having to read through the documents that are that are brought back and what could be improved in that in that process? And so I think, I think what’s what’s kind of super challenging sometimes is is when people don’t understand they have a problem until you alert to them them to the problem. And then once you alert them to the problem, they go like, oh my, oh my gosh, like this is massive. Like this is absolutely massive, and I need to address this immediately. But they weren’t waking up every morning saying this is a very specific problem that I need to solve.
Alexander Ferguson 24:56
So the major hurdle for getting to that early adopters is agitating the problem, making them realize that they actually there is an issue and a challenge with how they’re doing business currently.
Ryan Welsh 25:08
Yes, it is it is evangelizing the entire space. You know, like, like your, your outgoing hey, here’s, yeah, here’s the problems that you have that you don’t even know about. Here’s the, here’s the solutions that can address them. And here’s the ROI on those which, which I think takes focus. And so, you know, I think what a lot of AI companies either aren’t willing or aren’t capable of doing is focusing on a very specific business problem and solving, we’re delivering ROI to customers. And so they want to kind of stay in this abstract world of, you know, oh, we help you find answers in your unstructured text. Hey, that’s great. But what does that actually mean? Right, for who? Are you finding answers in the text? And why is that better than their current solution. And so it mean, even even at even at Kyndi, we’ve had to, you know, err on the side of being incredibly narrow in our in our focus and say, Hey, we’re going to help, you know, customer support people, we’re going to help, you know, internal IP support, we’re going to help these analysts over in supply chain, we’re going to help these analysts over in logistics, we’re going to help these analysts working in financial services that are trying to read these specific documents. And, you know, understand them the investor,
Alexander Ferguson 26:27
and you’re applying that specification or addressing that specific individual and use case, for messaging purposes, or for product purposes, or both.
Ryan Welsh 26:39
Both. Yeah, both. Because, because what, because what I think what, what you want to do is you want to deliver value to some specific group within the organization so that they can go to their boss, you know, say you’re you’re working with a director or senior director that, given the price point, they need sign off from a VP, like, they’re going to need to say, Hey, we can generate, we’re going to pay this company $250,000 a year, and they’re going to deliver us two and a half million dollars worth of value. Here’s exactly how they do that. Right. And as opposed to we’re gonna buy this for $250,000 a year, because there’s AI in it. You know, that? That doesn’t, that doesn’t work? Yeah.
Alexander Ferguson 27:26
Would you say that there has been an AI terminology overload in the past few years of everyone is saying AI this AI power that
Ryan Welsh 27:39
it is, it is the noisiest space in the world. It is, it is by far the noisiest space in the world. And I can’t wait to the day and I think it’s coming soon. Where we can actually stop saying we’re an AI company, right? Where you where you can just say like, Hey, we do this specific, specific thing. And there’s gonna be a handful of companies that that, that kind of keep the AI thing going, because they’re, they’re very deep tech, ai companies, you know, platform companies, but then there’s a lot of companies that I hope, that are just, you know, SaaS applications using a little bit of statistics that will just say, Hey, no, we’re a CRM solution, right? Because Because AI is in there, machine learning isn’t a differentiator. Everyone has it, you know, what is the what is the business thing that you’re actually doing? And I can’t wait to the day that we can actually drop machine learning or AI, and we can all just say, Hey, we all use it. Stop, stop saying it’s kind of like saying, like, you know, oh, I’m a software company. It’s like, yeah, we get it. Like, every you know, everyone is everyone that is a software company is a software company, like we get it?
Alexander Ferguson 28:50
What value are you doing? What problem? Are you actually actually solving your dress nicely, the getting over that first hurdle of just helping people understand. doing a search and coming back with links to your documents is not the same as how Google actually we’re starting to experience and as a consumer more often, I do a Google search. And at the very top, often if I’ve asked a very specific question, I get a snippet of my answer. Or if I do a google assistant, and it gets my answer, or Alexa or whatever, but putting that into a business environment, enterprise space, so just helping understand educating them. Is there still though a roadblock or a fear that arises of Will this actually work? Will will this roll out? Well, because it is still very new, and how do you overcome that fear?
Ryan Welsh 29:45
Yeah. We thought about this very early on. And we believe that that the people that can deliver value the fastest and demonstrate that it works the fastest are going to dominate the space So until what I, what I mean by that is, is that, in traditional software development, when you write code, it does the thing that you told it to do. Because you told it to do it, right. In AI, there’s this AI engine that cannot be, may or may not work. Like for the for the problem, like it may or may not like, hey, it works in 90% of the cases, and it may not work in 10% of the cases for whatever, whatever reason. And and so you’re consistently trying to answer the question, does the AI work for for your customers, and we think that the people that can actually prove that that works the fastest, are going to ultimately put systems into production and gain market share. And so and so you got,
Alexander Ferguson 30:49
when you say the proves that the fastest, you’re really saying of someone says, alright, show me and they say, okay, click this button. Now it shows you like the speed of execution from they are interested to it being deployed is that the speed that you’re so specifically
Ryan Welsh 31:06
working on third data, so here’s here’s the interesting thing about about machine learning is you can train it on a data set. And if the customer data is not even, like totally different, like actually just a little bit different, it won’t work on the customer data. And so now you’re saying to your customer, okay, customer, collect a bunch of data, send it out to either we’ll label it, or we’ll send it out to some other party to get it labeled, We’ll train our algorithms, and then we’ll come back in six months, and we’ll show you that it works. Right. And so like, six months, just to see that it works on your data is just way, way too slow. Whereas in traditional, you know, software, you know, you can just go in, and you can see that see that it works like oh, this, this works on my on my data on my structured data. Now, it’s it’s harder what you’re doing with with machine learning, but just all of that time and energy and effort that goes into proving that the API works. I think people just just lose interest. And they say, all right, am I gonna have to do this every single time? So say, my data isn’t static, say it updates? All right, well, how many times do I need to, you know, retune the models and label more data? And now we’re going through this process again? And it’s like, Is it always going to do this? And so yeah, I think it’s I think it’s, it’s both from from like, first meeting to deployment. But then there’s this crucial period in the middle, that’s first meeting just to proving that it works on the customer data that I think a lot of people fall down on.
Alexander Ferguson 32:40
I feel like, if that happened in the consumer world, consumers wouldn’t use a product if it if we assume I’ll sign up here, I press a button, and now I’m using it, it works. You’re and in the business world, it’s has to be somewhat as close to the similar as possible.
Ryan Welsh 32:57
Yeah, you try, you try to get there. And so and so I think the people that and so if we look at so for some of the more advanced AI platforms and use cases like like imagery and, and video and natural language, I think there’s a playbook that you can look at, in the data science space. So if you look at companies like like, data, robot data, IQ Domino data, lab, data, bricks, has a data science platform. So if you look at the companies that that are the leaders in that in that space, they developed auto ml. And so when you when you, you know, you upload your structured quantitative data, and you click on the variable that you want to predict, and you click on the variables that ultimately you want to use for the prediction, they’ll easily generate a model for you to use that is the best model to use. And you can see that value in a very short period of time. And so I think there’s a playbook for everyone else for the computer vision, image. nlp, nlp, platforms that if they can actually pull off with those data science companies have done in delivering value, the fast is relatively dominated dominated space. But the problem is, it’s much it’s much, much harder than than doing it on structured quantitative data.
Alexander Ferguson 34:22
Coming back to the initial challenge that you you recognize when when, when you were having to go through all those pieces of paper and finding those those items in the document like there’s got to be a better way. You’re painting a picture for a world where knowledge workers don’t need to do that, that the computer just is able to go through all this document to documents understand what’s the difference is what’s the future then for knowledge workers, what’s what’s it going to look like,
Ryan Welsh 34:53
spending all of your time on your most meaningful work. I mean, there’s nothing more frustrating than preparing data for the analysis that you need to do, like, what I really want to do is just solve this problem. But I’m going to spend the next three years collecting the data to solve the problem. Like, that’s just ridiculous. So the next several weeks, so the next several days. And so it’s kind of like, everyone has a research assistant. You know, when you think about it, and I think what’s interesting about what we do, compared to a lot of other AI companies is a lot of like very machine learning heavy companies are constantly trying to automate a process. So they’re trying to just remove people out of the process, let’s get this model to 99.9% accuracy. And let’s get rid of all the people. For us, we’re saying, hey, let’s have people in machines working together and make them more productive. And so we believe that that if we can, you know, help you find the answer to your question in a minute or two versus a day, well, then you now find the answer to your question, you can now go go solve the problem that you’re trying to solve for able to comb through 1000s of documents and extract all the relevant information that you’re trying to extract, you now have that information, and you can go Go do your problem. So it’s really trying to just remove everyone away from doing the mundane, low value work that can drive you crazy, to really just focusing on doing your most meaningful work.
Alexander Ferguson 36:22
What’s, what’s the new entry job going to be them? I mean, is it gonna stay the same? Because that would be the entry job, right? Hey, you go find all this information, go work on this stuff. But what’s the new entry job?
Ryan Welsh 36:32
Yeah, I think I think the entry job just just is, is well, not so much data processing, but it’s but it’s, you can now apply your skill sets your creativity and problem solving to actual problems. So I still think there’s going to be worked that needs to be done.
Alexander Ferguson 36:50
You just won’t be handler you’re working with you just handling with AI. Okay, this is what I got to work with them
Ryan Welsh 36:56
thoroughly. Because because things are constantly changing in the in the knowledge worker space. And so kind of unlike traditional manufacturing, where it’s like, hey, this thing does this specific task, every day, every minute of every day, every year from now into perpetuity. Like, you know, in knowledge work, it’s like, oh, we have this customer that came in with this with this problem. We have, you know, this person that came in with this problem, and you’re constantly like, maneuvering and so you need systems that can adapt with you can work with you to do your to do your most meaningful work. And I still think there’s there’s those entry level positions, but they’re not data entry. Right? They’re kind of maybe a little little higher up higher, Jane.
Alexander Ferguson 37:37
Okay, just just for fun. Everyone never thinks about AI. Often we’ll think about Iron Man’s Jarvis Riley good. Let me just talk to somebody that can go do the scans can do a bunch of research, come back to me with my answers. You know, how close can we get to that?
Ryan Welsh 37:53
No, I mean, not not very, I mean, yeah, I think I think I think a lot of people will say, Oh, yeah, no, I have a, I have a system now that does that. It’s like, get it get out of here. Right? It’s, it’s look, you know, we we want to enable people to more easily find the answers in large volumes of text, but it’s still up to the person to know what the answer. Right? And these are four very hard questions. So there’s simple questions like, you know, how, how tall is the Eiffel Tower? And you get an exact answer back? Or what’s the population of San Francisco? And you get an exact answer back. And you can kind of structure that stuff in a knowledge graph. And you can ask those very specific questions and get those specific answers back. But then there’s, there’s kind of these more complex, even more complex questions, just just different questions that you’re looking for this snippet of text. I mean, it could be anything from like, you know, how, how do I log into my computer, you know, it’s not working, like something simple that’s in a knowledge base. And and a lot of times, you’ll get like, links to a document that’s a user manual that then you then need to read through, like, wouldn’t it be great if is, if the doc if the system read through the manual and bring back brought back the exact sentences, this is how you log into your computer. But you know, there’s there’s not a lot of systems that can, that, that can that can do that. And so we’re pretty far away from from the Iron Man, Jarvis system, and it’s more about systems that just bring back more precise information for people.
Alexander Ferguson 39:29
I feel like if anything, we’re training ourselves to maybe ask better questions to start with is that another piece is like, well, I feel like in the old days of Google, you, you’d have to know the how to write the question properly. Or it’s just are we getting to the point where AI even if we ask it poorly, it can still understand.
Ryan Welsh 39:45
So what’s interesting about your about your question is I was I was talking to someone the other day, and they asked, how are you going to transition people to asking natural language questions because people are more familiar with, or more naturally inclined to write questions out there as they write them today. And I thought about that, and I was like, Wow, that is crazy. That the first time that I logged on to Yahoo, which was, you know, obviously, before Google. I wrote a question in the search box. It was like, I was asking a person, you know, like, Where’s the best restaurant in Philadelphia? Cuz I’m from Philadelphia. And, you know, you’d write that question. But it wasn’t until you learned that, hey, these systems don’t understand what I’m asking. You’re putting too much stuff in that question that you now right, Best Restaurant Philadelphia like? And what’s interesting is this individual thought that we are naturally inclined as people to ask questions that way. When we were reprogrammed, we were reprogrammed by keyword search to ask questions, questions that way. Now, he still has a valid point that there needs to be a transition because we’re now talking about 30 years of asking questions that way. And so we need to get back to asking questions in a in a natural language, way. But I just thought that was that was really interesting, just kind of how this individual thought that we were naturally inclined to ask these questions. And what’s interesting about you know, AI systems today is they can start to understand that context more in your question. And so language has as three things as syntax, semantics, and pragmatics. and machine learning is really good at understanding like the structure of sentences, it’s not necessarily good at understanding the semantics, the meaning of language. And it’s absolutely terrible at understanding the pragmatics of language, the intended meaning. And so, you know, if you can start to build systems that kind of understand all three of those those things, you know, people can ask questions in a certain way and get answers back that are basically what what your question intended to bring back,
Alexander Ferguson 41:55
our hotel and on on just your predictions of the future, in you know, 510 years from now, what do you predict? What will the it look like? What what are we going to see when it comes to natural language processing?
Ryan Welsh 42:08
And, yeah, it’s so I’m incredibly excited about all the breakthroughs that we’re seeing in natural language processing, and all the investment that we’re seeing in natural language processing. And all the use cases that are starting to be deployed and the focus. So we see like both the supply side of like, hey, algorithms are working, and we have these algorithms that can work. But we’re actually seeing the demand side as well, where, you know, we’re constantly getting inbound interest from large fortune 500 enterprises that want to transform their company into a natural language enabled company. And so when we actually think about, I think there’s some statistic like it’s 27% of employment costs globally, is for knowledge workers, even though they only represent about 9% of the global workforce. And so when you do the math, and you say, hey, if we can increase their productivity by 10%, then we’re adding $15 trillion in global equity market capitalization, which is just crazy to think about. And I think I think we’re, we’re at this point in time, where it’s kind of, you know, people reference a lot, but very similar to an industrial revolution, that kind of a knowledge revolution, where we can start to overcome our constraints, like I was mentioning earlier of only reading at a certain speed, only comprehending a certain amount of information as you read at a certain certain speed. And so I see this future that is just incredibly efficient, and incredibly valuable for all of us. That brings a lot of societal benefits with it.
Alexander Ferguson 43:45
Those that want to learn more about Kyndi. It’s ky n di.com. Let me just say that Kyndi, where’d the name come from?
Ryan Welsh 43:54
Yeah, this is really interesting. So my, my technical co founder and I were thinking about a name. We’re like, about to file the paperwork to incorporate. And I said, I said, Aaron, we got to come up with a name. He goes, I got it. It’s perfect. I got a perfect name. He goes, we’ll call it artificial intelligence thinker, Incorporated. And I was like, I was like, Aaron, I love you. But like, that’s a terrible, absolutely terrible name. And so I asked him good, because he had spent his early years, I think, was between like five and 10, or, or somewhere in West Africa. And I said, Did you have a name, a friend’s name or anything that sounded interesting when you said, Hey, I had a friend name, I think it was candy or candy or something along along those lines. And so we really, I kind of liked that. We played with it a little bit. And we found a, you know, just google the name and we saw Kyndi KYNDI, come back. And when you looked at it, it said someone with the capacity for thinking and reasoning, and we were like, Oh, there it is. There it is. It took us all 20 minutes to name the company. In Yeah,
Alexander Ferguson 45:02
I love I love the that simplicity of it coming together for that. Thank you so much Ryan for for sharing the journey that you’ve been on and the fascinating overall journey we are all on of natural language processing and where we’re headed. Thank you. Thank you. I will see you all on the next episode of UpTech Report. Have you seen a company using AI machine learning or other technology to transform the way we live, work and do business? Go to UpTech report.com and let us know