top of page
Indeed Wave.PNG
DS_logo_Primary.png

Talkin' LLM, AI, Parsing, and Tech.


Ya' down with LLM? How artificial intelligence is leveraging Large Language Models to power solutions like ChatGPT is redefining how recruiting and hiring is done. The problem is, its early innings, which means the masses don't quite understand the ramifications and how to harness the new tech in order to take their staffing strategies to a new level. That's why we invited Textkernel CEO Gerard Mulder on the podcast to make sense of it all. Textkernel’s LLM Parser, powered by GPT-3.5, represents a groundbreaking fusion of Textkernel’s proprietary industry knowledge with ChatGPT, making this dynamic combination the next generation in parsing technology, offering recruiters and HR professionals unparalleled levels of accuracy, efficiency, and insights into candidate data. Yeah, it's kind of a big deal, and the boys grill Mulder on the benefits, competitive landscape and potential threats. We also cover Textkernel's recent acquisition of Joboti, and dive into what potential moves the company might be eyeing in the future. Recorded live from the Textkernel booth in Paris at Unleash, you won't want to miss this conversation with long time Chad & Cheese sponsor.


PODCAST TRANSCRIPTION sponsored by:


Intro: Hide your kids. Lock the doors. You're listening to HR's most dangerous podcast. Chad Sowash and Joel Cheesman are here to punch the recruiting industry right where it hurts. Complete with breaking news, brash opinion, and loads of snark. Buckle up, boys and girls, it's time for The Chad and Cheese Podcast.


Joel: Oh yeah, what's up, everybody? It is your prime minister's favorite podcast aka The Chad and Cheese Podcast. I'm your co-host Joel Cheesman. Joined as always, the Jacques to my Cousteau, Chad Sowash...


Chad: Amen.


Joel: Is in the house. So we're recording live from Unleash World at the Textkernel booth...


Chad: Wow!


Joel: And what better first interview to have than CEO of Textkernel, Gerard...


[applause]


Chad: Gerard Mulder. [laughter]


Joel: Mulder joining us. Gerard, welcome to the podcast, for the third time?


Gerard Mulder: I think it's the third time. I did send you guys some recordings...


Chad: You get a... [laughter]


Gerard Mulder: So maybe it's four times? I don't know.


Joel: You get a red velvet smoking jacket.


Gerard Mulder: If he's named Jet, then I'm Gerard right now, right?


Joel: Gerard. Gerard. Gerard...


Chad: Gerard. Gerard.


Joel: A rose by another name smells just as sweet.


Chad: Like Textkernel, yes.


Joel: What do you prefer? Gerard, Gerard, Gerard?


Gerard Mulder: I'll do my official pronunciation on this show once and then I...


Joel: Give it to us.


Gerard Mulder: You can just... Gerard.


Joel: Gerard.


Chad: Gerard.


Joel: Yeah.


Chad: Gerard.


Joel: Gerard. Yeah. Alright, so you guys are making things happen. You're moving and shaking. Some news that recently came out. The Textkernel LLM Parser...


Chad: Oh my God, here we go!


Joel: Give us the skinny on this release.


Gerard Mulder: Yeah. So like with every development in machine learning over the past two decades, roughly we're also embracing large language models. And the great thing about these large language models is that the start that you have is already at a very high bar, it's better than anything that's come before. What the great thing about our capability now is that we're combining this large language model with our proprietary deep learning models. And by doing that, we can actually get it even more accurate, but also we're able to solve for a really big problem, which is the speed of the parsing. We're basically breaking up with our deep learning parser, the CV into bits, and then sending that into synchronous separate calls to a large language model. And this way, we can actually get to reasonable speeds still. I must say it's for our high volume customers. It's not the recommended solution right now because it's going to be too slow. But if accuracy is really important to you, it's great. And we're super excited.


Chad: Practical use. One of the things that cracks me up is we've been talking about large language models and ChatGPT this and Bard that and Claude, and we're not hearing real practical reasons to be using, literally, from a business standpoint. And this is probably one of the first. Other than going and putting your job descriptions into ChatGPT. It's just, it's one of those things where it's really cool to be, you guys are the market leader in parsing, matching all the way around. And now, instead of just staying still and waiting for the market to do something, you're doing it yourself.


Gerard Mulder: We were the first to release deep learning-based parsing engine. And now we're the first real supplier to deliver a large language model parsing engine. And we'll continue to also test different large language models. We're definitely going to jump on top of LLaMA, for instance, from Facebook, because... But the great thing they do...


Chad: Why not?


Gerard Mulder: Is they're making it open source, so that creates a lot of possibilities for us. That's great. And I believe that we're still scratching kind of like the surface of where AI or where digitalization and automation...


Chad: Well, you're just starting though. Right? You know what? I think most people don't understand is you don't just plug in an LLM and it just works at the max and best efficiency and/or capacity, right?


Gerard Mulder: No.


Chad: This is a learning type of prototyping, type of a thing. And so, I mean, it's going to get better over time, speed-wise, accuracy-wise, et cetera, et cetera, et cetera.


Gerard Mulder: Yeah. And what's most exciting about it actually, is kind of like where we're taking it next, which is sort of like, you have to imagine, you can ask anything, basically, you name it, we got it. It's actually a project we did during Innovation Week, the title of that, where we're using basically a large language model and we're letting you, kind of, on the fly, decide on the data model of your outputs. And where I think the value of large language models will come in is that you can combine what's in the profile of a candidate with the real world knowledge that's inside such a large language model. But the question is, how do you do that in an easy manner and also a scalable manner? We're figuring that out, basically.


Joel: And I appreciate Chad mentioning what you guys do 'cause we didn't ask at the beginning. We all just assume we know Textkernel and what you guys do. Would you add anything to the parsing side of it, in terms of a general view?


Gerard Mulder: Oh, totally. So, parsing, it is how we started. Just like Sovren did, which we acquired, definitely, with confidence, I can say we have the best matching engine in the industry. We have a sourcing application that help recruit our search easily on external sources and pulling the data into our ATS. We deliver the most comprehensive insight in the labor markets, demand side of the labor market through our jobs data products. Just to put a number out there, we've analyzed 1.6 billion jobs. And you can analyze that job history and job trends historically real-time based on scale. So for every company there is in the US or Western Europe, I know what skills they're hiring for, what was the first time they were looking for somebody that knows Hadoop.


Chad: You probably know better than they do.


Gerard Mulder: Yes, we do. Actually, we do.


Chad: Yeah. [laughter]


Gerard Mulder: No, it's so funny. It's funny that you say that, because I recently... We once had a customer, and they were talking about kind of like how to get more control on what they were writing in job descriptions and also, kind of like what everybody was hiring for. And I said like, "Don't you know?" They said, "No, we don't." And I said, "Well, let me show you. And I can also show you where it was published. And I can also show you the context, people that are mentioned in these contexts." And they were like, "Wow!" [laughter]


Joel: Long story short, we're biased, but Textkernel is awesome. Check them out if you need to. I want to get back to the LLM, the new technology. And so many times, a company in your position has to worry about disruption, has to worry about the next new company startup that shows up. How much of what you're doing is, we want to make sure that if we're disrupted, we disrupt ourselves? Is there a chance that this becomes the new foundation of Textkernel? Will this always just be a feature add? Is it a way to scare off the competition? Talk about the disruption and how you think about it.


Gerard Mulder: Yeah. Textkernel is very innovative, basically from the start. We have this competition we do called Innovation Week, where we actually promote bottom-up innovation. Everybody in the company can pitch an ID. And it's not like a hackathon. It's actually a one week where we close the entire company, and it's not just for developers, anybody participates in such teams. And so basically, in the spirit of that, and also a little bit inspired by Google, we started Textkernel Labs. And with these large language models, or you could say the capabilities...


Joel: I'm sorry, Textkernel what?


Gerard Mulder: Labs. Labs.


Joel: Labs. Okay...


Gerard Mulder: Labs.


Joel: Sorry, Labs. Got it. Alright.


Gerard Mulder: Yeah. Labs. Is my pronunciation not correct? Labs?


Joel: Labrador Retriever. Labs.


Gerard Mulder: Labs.


Joel: Got it. Okay.


Gerard Mulder: So what we're doing in Labs is building prototypes really quickly to solve problems that our customers sometimes come up with or create solutions that we think of ourselves, and we make them available to our customers for a particular portal. It's not for all customers, it's some customers that get access to it. They provide feedback. And when we say like, "Hey, this is actually really interesting and it solves a big enough problem," we then start to productize such things. So I think that's a really good way of starting to actually really test the waters with all the capabilities of large language models and generative AI in general, figure out, hey, what actually really works? Yes, I agree with you like you said, like a job description builder, how much value is it? Still, we had to build one, but we did it in two weeks.


Joel: People love the magic trick, Gerard...


Gerard Mulder: Yes, they do. They do.


Joel: They love the magic.


Chad: Pull the rabbit out of the hat, Gerard.


Gerard Mulder: That's in there, but like automated, reach out messages to candidates. Now, where we can add value in those types of things is because of our capabilities to analyze both a job description and a profile, and our kind of like our knowledge on the labor market, when I think about job descriptions, we know what your competitors are offering, what skills they're asking for. If you write a job description for a particular role, we can provide you that information and give you...


Chad: Because it's all public data.


Gerard Mulder: It's all public data. And we can give you suggestions. Actually, the job description builder was kind of like just a finger practice for us. But now customers are actually saying, "Wow, we want that, actually." And our partners are as well. So by building these things, we can also help our technology partners innovate faster at a very low cost.


Chad: So, you're like crowdsourcing prototypes at this point?


Gerard Mulder: Yeah, a bit. You could say that. Yeah.


Joel: Yeah. Innovation is a core part of your business but also more and more are acquisitions. Everyone who listens to our show knows about the Sovren acquisition, one that was very recent was Joboti and I think there was another one...


Chad: Excuse me?


Joel: Yeah. Excuse me...


Chad: Joboti?


Joel: Let me get a Kleenex on that.


Chad: So, talk about...


Gerard Mulder: I actually use that internally, Joboti... Yeah.


Joel: You used that? That's nice.


Gerard Mulder: So I said like, "Let's shake our booty and party, that goes... " [laughter]


Joel: I'm glad I'm making a positive impact on the world.


Chad: That's what you're giving to the world, Cheesman.


Gerard Mulder: Now we party like Joboti.


Joel: Big booty Latinas and Joboti, shaking booties everywhere. Talk about the acquisition strategy, the companies that you've acquired, what kind of companies are you looking for? Are these acqui-hires? Talk about that strategy.


Gerard Mulder: Yeah, so basically the strategy is two ways. On the one hand, we want to consolidate some part of the market because if we are on a bigger scale, then what we innovate and the innovations we bring to the market have more impact. So we can basically invest more in innovation that way. So that's one aspect of it. And the other aspect is actually acquire functionality that we haven't built ourselves, where we believe like, hey, this would really be a good add-on. So, Joboti or Joboti, well, it's actually pronounced, it's a very good example of that. If you look at Textkernel traditionally, we're building too many systems, like matching engines, parsing, data. But after the match, it stopped. We can automatically recommend great candidates to a job, but we don't control what happens after. Usually, not so much happened or it wasn't done in the right way. I wouldn't call Joboti the way we position it as a full blown candidate engagement platform. It's not, but it is the logical next step on what happens after you do a match or when a candidate comes to your recruitment side.


Chad: Yeah. Candidate data gathering and you guys are a data company, so therefore, you can, it's a better engagement to gather more data, and then you crunch more data and you give out better output.


Gerard Mulder: Yes, exactly. And like, hey, I found you a hundred good candidates automatically, now I've reached out to those candidates and 10 of them actually want to make an appointment, and we're making that appointment for you in your ATS as well automatically. That adds a lot of value. But what you are talking about is actually really interesting because the way I look at it like right now, many of our customers have huge databases. Some of them even with 20 million candidates in them.


Chad: Well, they've spent how much money building that database. Right?


Gerard Mulder: Unimaginable. Right?


Chad: Yes, yes!


Gerard Mulder: Yeah. Over years.


Joel: Unimaginable for Gerard is a lot, [laughter] just so everyone knows. That's a lot of money.


Gerard Mulder: Yeah. It's a lot of money. Yeah. But sometimes we do ask our customers, if we dare to be a little bit challenging, say like, "What is the value of your database actually? Or your people data?"


Chad: Yeah, that's a great question.


Gerard Mulder: In our Innovation Week, actually, we did a project where we started to create a dashboard and a dashboard would show kind of like what the potential value of a candidate database for a particular staffing company might be and what it is today based on the engagements of that company with their talent pool. And automation and the chatbots, and other like services around that, or analytics on your people data can significantly increase the value of the candidates database and can help you engage in a more relevant manner. And so, that's really where we're going and what we're building. Future acquisitions will also kind of like be in this path of trying to kind of like create a more automated flow and better communication with... Between companies and people, actually.


Joel: And one of the things I've always been impressed with is the moat that you've built at Textkernel and obviously buying Sovren really consolidated that moat and made it even more intimidating than it was before. But you do have competition, you do have companies trying to get stronger that have been around a while. I don't see a lot of startups. You can comment otherwise [laughter] if you do, but talk about the competitive landscape. How do you think about it? What do others have that you wish that you need to build better? Are they stronger? Are you just the 800-pound gorilla [laughter] and everyone else is just renting space in your jungle?


Chad: The big swinging gorilla.


Gerard Mulder: We're way too modest to ever say we are. Right?


Joel: You're way too Dutch.


Gerard Mulder: Way too Dutch. Way too Dutch.


Chad: Wait a minute, yeah. Come on. Come on.


Gerard Mulder: So here's the thing. This is kind of like our, has been our kind of like go-to market strategy. We're an API business and we deliver point solutions. Sometimes, APIs are point solutions as well but what I kind of mean with point solutions is say we build an integration into SuccessFactors to solve a sourcing problem and we actually build it but our API business usually gets integrated by suppliers ourselves. From the outside, it looks like some companies, I'm not going to name any company right now, looking at one over there, might be actually seen as a competitor, but they're also a customer. So we are...


[laughter]


Joel: I love when that happens.


Gerard Mulder: We are as a company, we believe very much in this composable product solution sets and what we're trying to do is give everybody in the industry, and I know, these are big words, we're trying to give everybody a leg up because we're trying to solve the hard problems that other companies can build on top of. When you look at competition, I have my traditional competition like XRI, for instance. And I appreciate them because we're kind of like the same age, started on the same bases. I'll never say any negative about them, but we have evolved beyond what we both started at and we're kind of like further ahead right now. We offer way more products [laughter] and solutions than they do.


Joel: Well, certainly competition is good for business. I don't think you'd want to be the only guy in town.


Gerard Mulder: Exactly. Exactly. And I think like I said, we're scratching the surface. There's so much possible. This market is still growing much faster than competition is coming into the market. And in the end, after that, it will consolidate, which could be a good thing for lots of people as well. Right?


Chad: Well, from the outside looking in, for many companies or many prospective clients, let's say, they see the parsing matching side of the house as there are many companies that are out there that do it, but they don't understand that you guys are actually the guts behind a lot of those other companies, some unicorns that say that they parse and match, and you guys are doing it behind the scenes.


Gerard Mulder: Yes. Best kept secret in HR tech.


Chad: Yes. Which is always, I always thought was genius. Right?


Joel: It's not a secret anymore. It's all on Chad and Cheese.


Chad: The white labelling... Yeah. The white labelling...


Gerard Mulder: Yeah. Well, we're not labelling anyone. Right? But yeah, you're right.


Chad: It's smart business for them though, because it's hard work doing... It's like one of the heaviest lifts in the industry is what you guys do. Parsing, matching, being able to contextualize data is fucking hard.


Gerard Mulder: Yeah. Building taxonomies, maintaining those. Yeah.


Chad: You know better than I do. Yes.


Gerard Mulder: Yeah. Ontologies. It's horrible. It's horrible. But we love it.


Chad: It's horrible! [laughter] It's horrible. But we love... I wake up in midnight sweats all the time, but I love it.


Joel: If it was easy, everyone would do it. Right? We talked about partners, competition, startups. We're here at Unleash. What are some of your takeaways of the conference? What are some of your goals here by exhibiting? What do you hope to get out of this week in Paris?


Gerard Mulder: Yeah, regrettably, I'm only here for one day. It's just two days. Primarily happy I'm talking to you guys, of course.


Chad: Thanks for being here, Gerard.


Gerard Mulder: And having you on our booth. So that's cool. Yeah. For me, it really is a networking event and it's just like, and I just like, interesting company right behind our booth that's... I forgot their name, but... [laughter] Sorry about that. It's just like a new company and we're just talking about kind of like what are their biggest hurdles in development? And that's actually, skills and ontology. And they're like, it's such a big lift for them, but they have a great product to help people progress their careers. And just like in a regular conversation, you suddenly have a new potential new partner that will use your product. So for me, it's networking, just understanding what people are building...


Joel: I think he's going shopping is what I'm hearing, I'm hearing.


Chad: Yeah. If you take a look at it, all the startups that are around here, if they need data and 99.9% of them need data, they need somebody to parse it, contextualize it. It's like, this is shopping from a couple of different aspects...


Joel: It is dating, yes.


Chad: Number one, new clients in the prospect of, who knows, M&A one day.


Gerard Mulder: Yeah. The other day, like many of the exhibitors here, are somehow using a software component of ours. The other day, some time ago, I was at the Borne event and I didn't expect it, I felt like, okay, this might be more competitive actually. They're more focused on staffing, of course. But even there, I suddenly noticed, probably 30% of the suppliers here use either like Sovren or Textkernel type of technology. We're trying to be nice to everyone basically, of course... [laughter]


Chad: Smart.


Gerard Mulder: And help out where we can.


Joel: Yeah. Your brand is not one of animosity and fear...


Gerard Mulder: No.


[laughter]


Joel: It's a loving touch that I get here...


Chad: It's the red light district of the HR tech.


[laughter]


Gerard Mulder: Okay. Yeah...


Joel: Everybody's wel... It's like, oh, whoa, whoa, whoa...


Chad: It's very inviting!


Joel: Now you've gone too far, Sowash.


Chad: It's very inviting.


Joel: Now you've gone too far.


Gerard Mulder: No, but we'll take anybody on our booths, on our Textkernel booths. Whether you're a competitor or partner or a customer, we don't care. We'll take you around and you'll see.


Joel: That's right. There's an electric bike and a pair of wooden shoes for everybody...


Gerard Mulder: Yes.


[laughter]


Joel: At Textkernel. Gerard, thanks for your time. Thanks for letting us camp out here in your booth...


Chad: Yes.


Gerard Mulder: Yeah, you're welcome.


Joel: For all of our listeners that don't know you, that want to connect and learn more, where do you send them?


Gerard Mulder: Www.textkernel.com, of course, and you can also hit me on LinkedIn, Gerard Mulder. It should be easy to find in Amsterdam.


Joel: Absolutely. And the best coffee is here at the Textkernel booth, by the way, as well. Chad, that's another one in the can. We out!


Chad: We out!


Outro: Wow. Look at you. You made it through an entire episode of The Chad and Cheese Podcast. Or maybe you cheated and fast forwarded to the end. Either way, there's no doubt you wish you had that time back, valuable time you could have used to buy a nutritious meal at Taco Bell, enjoy a pour of your favorite whiskey, or just watch big booty Latinas and bug fights on TikTok. No, you hung out with these two chuckleheads instead. Now, go take a shower and wash off all the guilt, but save some soap because you'll be back. Like an awful train wreck, you can't look away. And like Chad's favorite Western, you can't quit them either. We out.

Comments


bottom of page