224: Trending Topics - The REAL Impact of Natural Language Search ft. Ilia Cheishvili

Description

In this episode, Sam chats with Loxo's CTO, Ilia Cheishvili, about something hyper-relevant: Natural Language Search. In case you missed it, Loxo just launched NLS — and it's already saving recruiters so much time and mental energy.

Throughout the conversation, they explore how NLS allows recruiters to source candidates more intuitively, moving away from traditional Boolean searches that can exclude potential talent. The real kicker? It's becoming increasingly important for recruiters to acknowledge the importance of flexibility in candidate requirements — which leads to the need to cast a wider net in their initial sourcing efforts. As someone who has built the product from the ground up, Ilia shares insights on how NLS can surface the most relevant candidates and improve hiring outcomes — while also addressing common misconceptions about candidate filtering.

Long story short: If you've been curious about all the buzz around NLS but need a bit of convincing before giving it a try yourself, you've come to the right place! 

Transcript

Sam Kuehnle (00:00.948)
Hey everyone, it's Sam Keenly and welcome back to Becoming a Hiring Machine. This is the show dedicated to fixing recruitment by going beyond saying what needs to change and instead teaches you how to make that change. Today, we've got a great interview ahead of us, but before we get into that, I want to tell you a little about the show. Essentially we have shows within the show here. Some days like today, we have interviews with industry thought leaders or our own CTO who are shaking up the space. Other days we cover trending topics.

Ilia Cheishvili (00:11.182)
Thank

Sam Kuehnle (00:25.748)
Stop by every Tuesday for tactical Tuesday episode where we go deep on how to do something. It's going to help you drive better results in your day to day. And occasionally you get a mic drop episode from Matt, where he shares something that he's been thinking about within the recruitment space and wants you to know. As a reminder, if you the content that we share here in the show, make sure that you subscribe for more. All right. Today's interview recruiting has long relied on a specific kind of knowledge, Boolean strings and rigid filters. So while these are powerful,

Ilia Cheishvili (00:39.946)
Thank

We're going to stress it out.

Sam Kuehnle (00:52.872)
These methods are often slow to master and can accidentally exclude great candidates. They force you to think like a machine translating human nuance into complex query. But what if you could just source like a human instead? So natural language search, you'll hear us reference the acronym out NLS over the course of this conversation is exactly that. It's an AI powered tool that understands your intent, not just the keywords that you're using. And that allows you to describe the ideal candidate in plain language, just like you would to a colleague or a friend.

Ilia Cheishvili (01:04.91)
Thanks

Sam Kuehnle (01:20.808)
And this is something that many recruiters are beginning to adopt. So how does it work? What's the best way to get the result that we're looking for? So instead of me answering these questions for you, we brought in our CTO, Ilya, the guru himself to go through this one together. So without further ado, we'll get into it in today's episode with Ilya. So Ilya, welcome back. Yeah. You get the fun conversations you were here last year, I think it was email deliverability and you're just like, my God, this beast. So.

Ilia Cheishvili (01:40.044)
Thank you. It's always good to be here.

Ilia Cheishvili (01:46.348)
Yeah. Exactly.

Sam Kuehnle (01:48.244)
Yeah, just another very small nuance to one in the world of AI and natural language search.

Ilia Cheishvili (01:52.185)
Well, you know, it's what I tell people because this question comes up a lot and I say, look, just imagine a six year old boy staring up at the clouds and thinking to himself someday, I'm going to grow up and deal with a lot of compliance. It's the dream, right?

Sam Kuehnle (02:04.052)
Yeah, our two year old right now she's torn between being a what she wants to be either a trash truck driver or a pilot, you know, so maybe we'll get compliance on that list for her shortly after this.

Ilia Cheishvili (02:14.99)
Yeah, would definitely be a trash truck driver. It's so exciting. How often do you get to fly and then it's just like you're in different city all the time, you're on the road. That's not fun. Yeah. Yeah.

Sam Kuehnle (02:18.26)
and so I'm leaning towards, right? You can drive around all day, windows down, smells great.

Sam Kuehnle (02:26.79)
No, no, we'll do it. So natural language search. This is newer. You know, it's, really coming up against the traditional Boolean and filters. just like starting at that level, how is natural language search different from these two?

Ilia Cheishvili (02:34.743)
Mm-hmm.

Ilia Cheishvili (02:42.734)
Yeah, well, it's kind of like you used to have to brush your horse and feed it and, you know, change the horseshoes out and do all that stuff. And then we got cars and you kind of just sit there and mostly goes where you want it to go. And then after a while, we got this thing called like full self-driving. I know it's still supervised and it's still coming soon. But the idea is, you know, like I don't want to do that much work to get kind of like, we know what the outcome is going to be. If I'm hiring an account executive or a certified public accountant, we kind of really know when I throw that dart at the board what I'm trying to hit. So the idea with natural language search is I want to take a lot of the drudgery out. Like, don't make me think. We all know what we're both looking for. Let's just go and find it, right? Like, help me expose the most relevant people that I should reach out to.

Sam Kuehnle (03:34.164)
So why are you so bullish on it?

Ilia Cheishvili (03:36.462)
I think I'm bullish on it for the same reason that I was really bullish on when, for example, we got, I'll pick on Apple, right? And not even pick on like kudos to them because they took iPad OS and put that into head into the hands of, you know, every child in America, so to speak. if you know how that goes, right, we're both parents and, it's something that we're computing and device usage used to be. I'm not going to say it was a priesthood, but you know, you've had to

Get the desktop, get the monitor, plug in the keyboard and mouse and you have to know how to install drivers on Windows. Everybody remembers those days if you're a certain age and they made that more accessible. They made it so that you get good outcomes without having to like think so hard, right? Because I mean, you know, I want to go and browse the web. I want to play some games, use some apps, doom scroll, right? Like I shouldn't have to understand how to install an Nvidia GPU just so I can doom scroll. And so.

natural language search is kind of like that, right? It makes the whole thing way easier to use and more accessible. And if you are trying to find one role that's really hard for us, you know, took us a while to fill with our distributed systems engineer. It's pretty neat. There's certain things and backgrounds and skills and whatnot that you have to target and how are you going to keep 200 things in your head and not miss something. And it's a very low hanging piece of fruit in hindsight that

Sam Kuehnle (04:39.902)
Mm-hmm.

Ilia Cheishvili (05:02.155)
just come and help me with this is with really the idea, right? Is how can you make this easier and have a great outcome? Because it's just so fragmented the way that people approach these things right now.

Sam Kuehnle (05:04.083)
Yeah.

Sam Kuehnle (05:13.906)
Yep. So let's do, let's do this. Let's go through an example. There are, and we're not going to be naive and say, we're the first, first to market with the natural language search. There are other tools out there that have natural language search. Yeah. And so that's, we'll use an example and that's also as we're going through it, we'll cover kind of what's current state for these early adopters who are using this and other tools. And then we'll talk about how it operates.

Ilia Cheishvili (05:22.701)
You don't want to be the first to market with yet. Yeah.

Sam Kuehnle (05:42.601)
with Loxo and the future that we're painting. Cause there are some, different nuances there as well. And that not every natural language search is the same as you go from the different platforms. So not only are you learning, you know, Boolean is Boolean is Boolean, but NLS is not NLS is not NLS. So, you shared yesterday in our, in our internal, Slack channel for the team, a fun example, you said you were looking for someone who

has FinTech and SAS experience, and also has a PhD from Cambridge.

Ilia Cheishvili (06:15.863)
Well, it was funny because the team came at me, right? And they were like, look, I looked, I use natural language search, right? And I'm looking for somebody in software from Cambridge has a PhD from Cambridge, right? And they're in Fentuk. Why am I getting so many results back? It should be fewer than this. And I'm like, well, guys, you know, we actually did our due diligence here. The product team, the engineering team, I mean, we went back and forth. If everybody looked,

We all know we have, I have a red book on my bookshelf called Death by Meeting and this was worse than that. But it's what we have to do to get to a good outcome. Now the reason is most people, think of it incorrectly, right? They're like, yeah, I really want that like suffer a PhD from Cambridge that's done finance. I'm like, that's, that's actually two people. We tried that search, right? There's like very few people that have done that. And chances are, even if you can get them interested, Zuckerberg is going to come in with like a hundred million dollars. You can't get them anyway, right? Like, so they're off the market right away.

Sam Kuehnle (07:06.558)
you

Ilia Cheishvili (07:11.179)
We see this happen with lot of recruiters and hiring managers. come in with, know, especially if they're not very experienced, they'll come in with their list of requirements. Now, if everything is actually a requirement, that person doesn't exist. immediately what happens is you get certain requirements getting relaxed. well, you know, yeah, you know, those guys from Oxford, they can't even eat a pizza the right way turned around. But yeah, we have to settle. We have to go from somewhere from like Oxford, Harvard or Stanford. Like,

Sam Kuehnle (07:19.046)
Mm-hmm.

Ilia Cheishvili (07:37.966)
I really want Cambridge, like we just have to go to like the C tier, right? That's how people behave themselves. And yeah, the interesting thing is that you learn that hard requirements actually aren't and almost everything is a nice to have, right? Like, so what's an example of an actual hard requirement? It's like, look, I'm not willing to sponsor a visa. Okay, I get that. Well, then that does actually cut people out. But for everything else, what you're trying to do is you're trying to get as many of those things as possible.

And those are the people you want to see ranked first because we don't want people to have this experience where you come in and you describe what you're looking for and then you get nothing, right? For people who are going to take Aloxo as like not a good tool. It's actually not the truth. The right way to present it is to say, what is the best possible fit? Because you're in the market, you want to hire someone, right? But if someone doesn't have the exact shade of green colored hair that you want, so are you just going to not grow your business? Are you not going to make a hire, leave money on the table? That's not a good option, right?

Sam Kuehnle (08:33.481)
Mm-hmm.

Ilia Cheishvili (08:35.563)
And so a lot of it is actually, kind of get educated as to what the market is when you come in and start using natural language search, right? You're like, okay, I am out of millions of people. get the very best at the top. You're probably not going to go to page 100. I know some people do that on Google, but almost nobody does, right? If people are going to look through maybe the first couple of pages and the higher that you want is going to be in there. Now, maybe they're in the market.

Sam Kuehnle (08:52.978)
Mm-hmm.

Ilia Cheishvili (09:01.237)
or not in the market, maybe they want too much money, then it gets into that kind of minutiae and then you would work your way down the list. But part of the educational process is basically saying like, yeah, I really wanted all those things. They don't exist, right? Here's the next very best possible thing. okay, I wanted that PhD in software finance from Cambridge, but here's someone from Caltech, probably a really good fit, right? I know a couple of guys from Caltech. They can...

They can microwave Totino's pizza rolls without exploding something. They got that high IQ. They're going to be a good fit for your organization. You don't have to just pin it so hard to Cambridge. And that was a great example because it came up from one of our account execs. Trying to understand how it works. Would really like the best way to summarize it is, I want to see the most relevant candidates at the top. I don't want to see no results, which is what would happen if you were so strict.

Sam Kuehnle (09:30.676)
Ha ha ha ha ha ha ha ha ha ha ha

Sam Kuehnle (09:43.262)
Yeah.

Sam Kuehnle (09:51.771)
Mm-hmm. that's what I think one of the biggest differences in the, lot of recruiters are used to from legacy sourcing tools and even some of the other new NLS tools is that they're all honing in on what's the overlap of requirement one, two, three, and four. And so it's that perfect center of the Venn diagram. Whereas, correct me if I'm wrong, we're giving you the full four circles of the Venn diagram, but we're putting the most relevant people.

at the top. And so you get to see the full global universe of everyone that fits maybe just one of the requirements instead of trying to find there's only two people in the world and then having that be the sole output.

Ilia Cheishvili (10:30.22)
Right, exactly. Yeah, you want to have the highest probability of a successful outcome, if that makes sense. And cutting too many people out and being super strict, you're not going to be successful that way.

Sam Kuehnle (10:38.964)
Mm hmm. So.

Sam Kuehnle (10:44.912)
Yeah. And let's talk about that part a little bit more. Cause I think that's going to be the part that so many are hung up on because that's just what they're used to. They're I'm as I'm fine tuning on this. want to get to my candidate pool where there's only 250 people. So I can go and make my shortlist off of that. This is a change in how you approach this. So instead of viewing it as like, I have to get this pool down small. How are you approaching this? Are you just saying, you know,

Ilia Cheishvili (10:50.252)
Yeah.

Sam Kuehnle (11:11.816)
first five pages or your first 250 people use it from that sense or like how are you thinking about this if you're going into doing this search?

Ilia Cheishvili (11:19.02)
Well, that's exactly right. So don't get intimidated by the count, right? It's basically you have the most relevant on top and then they get less relevant as you go from page to page. If you're used to Google, you're used to this. And don't think that like, okay, we returned a million results, but you don't have to go all million, right? And people are used to that. You, we actually give the option. I would say start at the top and work your way down. That's the best. That's how I tackle lists myself. It's, it's generally good advice. But then the other thing is,

Sam Kuehnle (11:28.754)
Mm-hmm. Yeah.

Ilia Cheishvili (11:48.685)
It's not like you have to stop there because we kept the traditional bullying on the left. So if you really are like, hey, man, you know, like I care really strongly. We only want University of Alabama. That's where I went to school. That's where all the cool kids are coming from. Go ahead and filter for it. Right. You know, I probably advise against don't cut it down too much because then your pool is smaller. And then sometimes you are down to like 10 people and they're not in the market, but in general, top to bottom. Right.

Sam Kuehnle (11:53.972)
Mm-hmm.

Sam Kuehnle (12:14.856)
Mm-hmm.

Yeah. And so it sounds...

Ilia Cheishvili (12:18.758)
And don't think like, man, if I don't make it to page 900 in my phone, where are going to before with traditional search?

Sam Kuehnle (12:25.788)
I hope not. Otherwise we're doing something wrong. So it sounds like then, yeah. So your recommendation for how to use, how to use this and where we're seeing the vision is type in your query into the natural language search part. That's going to get you this full universe. And that's going to have all your real and secondary requirements, not necessarily the cover ones, but it's going to build your universe. But then if you do have the hard and fast, true requirements, they need to live in London.

Ilia Cheishvili (12:28.04)
Yeah.

Ilia Cheishvili (12:41.236)
Okay.

Sam Kuehnle (12:54.088)
they need to have Python, you know, any of that. That's what you're saying. Okay. That start using Boolean or filters to hone in and tighten that up a little bit more at that point.

Ilia Cheishvili (13:05.036)
Well, I mean, exactly. It really is a mindset shift because people have been trained for so long to say like, I need to keep cutting and cutting and cutting until this gets down to 300. I'm like, well, the bad news is just inevitably, I mean, we're not perfect. And I just mentioned, right, like when we talk about natural language search and, know, one thing that makes us different in lock. So we actually train our own models, right? But I guarantee you, and you know, there's implications to me naming names, but pretty much everybody else is just like, they're like, hey, open AI help like.

Could we just talk to your model? We train our own. Yeah. We can use the best of breed of everything if we don't train our own. So, that's where we have the advantage, right? Because as you're doing the traditional bullying, bullying and filtering, here's the problem. Like, let's say you're hiring somebody and you know, I'll use something that's relevant to me as like somebody who is really good in Python. You just mentioned it, right? Some people call themselves like Python ninjas.

Our model has been trained to understand that you by just traditional bullying filtering will miss that person. That person might have been the perfect hire. And which is why again, like natural language search will just surface the most relevant to the top. You won't miss them, right? The problem with people right now with their mentality is they're like, I need to keep slicing at this list until it's down to 300. I'm like, probably in position number seven would have been that Python Ninja, but you're missing that person. You're leaving money on the table. You know, that person could be working a better job.

None of that can happen because we're thinking in this old school way of like, need to remove results. No, you want the most relevant to show up on top.

Sam Kuehnle (14:36.627)
Mm-hmm.

Yeah, and in a way we kind of are removing the results. We're just, we're not removing them. We're just lowering them on the list. They're still there, but yeah.

Ilia Cheishvili (14:43.5)
Yeah. Well, yeah. And that's the point, right? It's like, well, maybe you're, cranking through people and like, you know, it's just not working out for it. Maybe you make it to page 10. I mean, that's better. It's better to be able to make it to page 10 to a like more relevant candidate than to never see them at all because you typed in, you know, Python developer and you missed everything else.

Sam Kuehnle (15:01.342)
Mm-hmm.

Yeah. And this is a topic, this, I think this episode will be out shortly after this one. Vivian and I did for tactical Tuesday where it's assumptions kill a lot of searches where you think, you know, you might have someone like, Oh, they're not going to take this pay cut. So you rule them out or any specific seniority above this. Well, maybe they've changed their life priorities or, know, it could be any other instance like that, but this helps avoid a lot of those assumptions where you're going to rule out potentially phenomenal fit candidates due to some of this. now they're going to be surfaced for you.

Ilia Cheishvili (15:12.406)
Jeff.

Ilia Cheishvili (15:17.738)
Right. Right.

Ilia Cheishvili (15:33.1)
Totally. Yeah, exactly. And if you actually hear to our CEO, Matt, the way he always talks about is you always want to cast the widest possible net, right? And then work your way from there. And that was really kind of what we had in mind. by the way, I'll tell you, I came at this initially with a little bit of a different approach and we were cutting people out and we sat like, why did it take us so long? Why were we like, you know, getting shredded in meetings? We were looking at results and we were saying, okay,

Sam Kuehnle (15:52.212)
Mm-hmm.

Ilia Cheishvili (16:00.681)
here's a version of natural language search that takes the pool of candidates down to 50, but that's not good, right? So what if it's not enough? We're missing a lot of stuff if we do that. And so then we said, no, no, no, it would be better to have the most relevant on top and let people work their way through it.

Sam Kuehnle (16:16.82)
Yep. Okay. I love it. So yeah, this has been a fun one. You guys have been heads down on this for a while and it was a big announcement and exciting announcement. It's, say, NLS is new to this space and even within it, it's going to vary from product to product. So that's why I wanted to chat with you a little bit, just to explain so people understand like, what is this new sourcing methodology that people are using? Why is it different when I try it in this platform versus Loxo or anything else?

Ilia Cheishvili (16:18.057)
Yeah.

Ilia Cheishvili (16:39.499)
Well, and one of the things that I just love talking about this a lot because it's not like we're done, right? Some of the things that we're working on right now are they're not happening anywhere else and they probably won't. And what I mean by that is the amount of intelligence and the kind of models that we're training. You'll hear, for example, OpenAI, Anthropic, Google, DeepMind, they use this whole terminology of like mixture of experts. That means a lot of things.

Way too much to get into on the phone right now, but from where we are and from where we're going, just the quality of results, right? And how they keep improving and compounding on themselves. I actually, I'll quote myself from a call an hour ago where I was saying, you know, if you're a well-educated customer and you actually understand and have tried everything, I'd use Lockso to source and hire my team. And I've tried all the other things, right?

Obviously being in the space I get retargeted on Instagram and on TikTok. We all know who's advertising on there right now. You guys all scroll, I don't even have to name it. It's just not as good. And what I love about it, like I'm a competitive guy. I love watching that gap open up more and more aggressively. And we're going to continue to see that happening, that compounding effect, right? We all know what a small step every day you'll find yourself in a different field after a while.

Sam Kuehnle (18:02.196)
Mm-hmm. Yeah, I love it. I think it's...

Ilia Cheishvili (18:03.883)
And then wish I'd get into details, but man, proprietary stuff, know. Yeah, exactly.

Sam Kuehnle (18:07.54)
Stay tuned for the next super fun episode with Ilya where we can turn back even more technical things. Any final thoughts or parting things that we didn't touch on during this that you think would be helpful?

Ilia Cheishvili (18:11.815)
Yeah.

Ilia Cheishvili (18:20.191)
Yeah, I would say the big thing has been a mindset shift, right? Instead of saying like, have to really come at this with an axe and chop it down and whittle it myself. It's to really say like, you know, trust the process, work through the results, like the best stuff's on, best off the most relevant stuff, it's on top, right? And so it's a big trust building exercise.

Sam Kuehnle (18:40.019)
Mm-hmm.

Sam Kuehnle (18:45.682)
Yeah. Okay.

Ilia Cheishvili (18:46.315)
And it always comes down to, Like tech, it's what's the word that we use for that? It's incidental, right? What we do most of all is try to build that trust that like, yes, this is a better way.

Sam Kuehnle (18:59.159)
Mm-hmm Yeah, yeah. Well, I'm excited to see where this goes. I know we we eat our own dog food We do this all the time is where we're growing our team and hopefully we get some some good stories from the rest of the market So Ilya as always I appreciate you coming on not just to share the technical knowledge But your your metaphors and analogies will forever be undefeated So I'm glad you glad you could hop on with this

Ilia Cheishvili (19:06.142)
Okay.

Ilia Cheishvili (19:20.223)
I know, right? That's with zero prep. Can you imagine? Yeah. Exactly. All right.

Sam Kuehnle (19:26.698)
challenge accepted. Next time, Ilia, you have three days to come up with all the metaphors possible for this. All right.

Ilia Cheishvili (19:32.445)
Okay, accepted indeed. it's always good seeing you on here.

Sam Kuehnle (19:36.816)
Likewise likewise well as we wrap becoming a hiring machine is a production of loxo if you like this episode You're not sure where to go next we've got a link to a similar one and you can find that in the show notes or on our website loxo.co Slash podcasts full video starts are available on YouTube if you want to see our beautiful faces there if you ever have questions somewhere way podcast loxo.co and Lastly if you did enjoy this episode, please make sure to leave us a quick rating in your podcast during platform takes you three seconds That's gonna help others like you to find us. So

All right, everyone, that's the show. That's another fun interview with Ilia. Until next time, bye, all.