Ignite Startups: Prachie Banthia on Reinventing Technical Hiring with Lightscreen AI | #132

0:00 this is like advice that I've heard from other people and I'll just reshare it because I think it was really helpful to us is if you're like pulling teeth to get customers to use your product you're probably not finding product Market I've been in this place I'm like oh no it's because I'm not following up correctly my messaging is bad like I'm not using the right channels but like if people aren't coming it's probably they don't

0:18 want it versus in this case finally we had figured out something where we were like oh people want to like people are ready like today to start using this can you can I use it now like how much does it cost yeah so those questions like 15 minutes into the first call is when I was like oh this is like a real thing this is a hair and fire problem that people want they have a budget to solve it right now there's a

0:39 real cost hey everyone welcome back to the ignite podcast today we're delighted to have prachi bantia founder and CEO of light screen AI PRI has an impressive track record in the tech world having LED product teams at assembly Ai gopuff and even started her career at Google as part of the prestigious APM program wow with her expertise in AI product development and Marketplace growth she's now Building light screen AI a YC back Cutting Edge platform transforming

1:04 technical hiring really cool demo actually I'm wonder if we'll get into that um let's dive into our journey um welcome to the program thanks for having me excited to be here yeah it's good to good to see you again it's been a couple months since we met at demo day for YC congrats on all the progress I know that's always an arduous Journey coming coming through that program oh yeah the journey is just beginning yeah exactly

1:25 our journey continues but oh yeah it was great so I mean you had a great career going like what inspires you uh to switch lanes and and go into kind of a Founder mode I think I'm probably like a lot of people and that the entrepreneurship bug was always there small size when I was in college I with the friend started a dance company and so that happened and that like has still that company still exists and like it

1:51 gives me so much joy to see instram and they have no idea who I am which just like feels amazing that they like you know are just doing their own thing now it's an autonomous thing and then when I uh was at Ross which was a small startup that uh we sold to go puff I just worked harder there than I'd ever had in my entire life and I was kind of shocked at myself at how hard I was working and it

2:09 was because I cared so much about the people I was working with and the the like Mission we were trying to solve for and I was like if I could do this but with the risk profile of a Founder like work this hard and care this much it just seems like amazing so I kept getting closer and closer to being a Founder until eventually this moment in time I'm sure everyone's listening to your podcast like broken record to them

2:29 but it is such a moment in time to start a company like there's just so much exciting stuff happening everything is changing you can start a company I feel like in anything that you're interested in and still and be able to upend it's like mobile 2009 right now is what it feels like because I kind of looked through that last iteration I wasn't quite in Tech yet that that was about the time I started getting into Tech was

2:48 2009 10 so it kind of has that energy to it where you're like yeah there's just there's so much to do I just feel like you could do you could go anywhere and there's something interesting to build which I feel like was not there even 5 years years ago Gavin my co-founder and I had known each other for a decade and he had wanted to start something I was finally in a place like just like financially and mentally where I was

3:07 like stable enough where I was like I can like do something more fun and it was such a perfect time to do it that I had to just make it job yeah how how did you guys meet I mean that's pretty unique to know your co-founder for 10 years or more so I um so I met Gavin officially when I was working at Google Gavin and I worked in the same building on the Google Cloud team we didn't work

3:25 like professionally very closely but I do think one of the great things about someone once told me that when I was starting at Google or deciding to work at Google that one of the best things you get out of Google is just like you get to know a lot of great engineers and that might be your future co-founders one people yeah Gavin yeah and I met at Google but we weren't really working together we just like would play board

3:42 games together and things like that and I ended up marrying his best friend and so when my husband and I were dating I was basically living at Gavin's house because they were roommates and so I got to know Gavin very well just like personally during that time and when I was optimizing for who to start a company with it's of course the like you want a 10x engineer who you can trust do the engineering work but I think just as

4:01 important is someone that you already know you like and like you could spend random Wednesday with and it'll be like a little bit more fun you know and you can like go through the hard stuff with so I did optimize also for just like values and someone I like want to spend a lot of time with yeah and I think that's something that YC looks at pretty closely in their application processes what's the history of these co-found the

4:21 there are founder teams there that like just met right before YC for sure but I think it which I who knows to them like I feel like that would be hard for me to like invest so much time with someone I barely know it's rare definitely really nice if if the people have known each other a long time because I think founder breakups are one of the biggest reasons companies fail out of YC so they just don't want that like you know

4:41 they're trying to minimize failure as much as I I see it all the time in my portfolio I mean of the shutdowns that we've had I can think right off top of my head yeah like two or three of those are one of the founder leaves and yeah actually three and I think two of them are YC companies actually or one of the founder leaves they just decide it's not for them anymore or whatever they're go

4:59 back to their cushy s figure job at Google or whatever yeah I think grit there's a great book uh called grit which I would recommend to any founder it's an older book I guess um but it really left an impression on me at just it's it's a lot of Doon as a talk at the end of YC of just like just don't fail and that's like half the about or just don't give up and that's don't give up

5:17 battle if you're already like decent at the other stuff like just keep trying things and I think was something I really respect a lot in Gavin Gavin has wanted to do a start up for a long time and it's kind of not given up on that dream and I just know it's someone who's not going to give up which I love it you you know I think having accountability of a partner is helpful that's amazing so how what was the origin story behind

5:36 light screen so let me explain light screen yeah thanks Brian so light screen is an uh voice and video based AI interviewer um we're already being used by I guess many multiple companies today to uh screen their technical candidates so it's an AI it's like a drop in replacement for a coding interview and the big differentiator is we don't just assess coding skills because we actually think that's not going to be what is the most important for an engineer in the

6:01 next like five years but actually the critical thinking skills problem solving technical communication at scale and the idea came so I'll say a few things so one is we did have a bit of like you know the drunken walk is I guess what people call it or the random walk of you know you're trying one idea you're prototyping it you realize it's a terrible idea you try another idea you're prototyping prototyping helps you realize what is actually a good idea and

6:23 what is it so we did a lot of building I have some friends from business school who are also in the founder we're so like try to start something and people can get into like analysis paralysis so instead we our orientation was just like let's build something shitty and like see if we feel like this actually would be good um and so we did a bit of that and also G and I just spent a lot of

6:43 time brainstorming like you know we know AI is going to change everything especially the tasks that are most low Roi repetitive expensive for companies today and what are the most low Roi repetitive expensive tasks that we had to do or our teams had to do at our previous jobs that was doing technical interviews and so we started first with a actually something kind of less ambitious like a co-pilot in your interview that is sitting there taking

7:05 the notes for you and doing the evaluation of the candidate after the interview like of a of a human to human interview but when we built it we were like this is fine but there's actually something way more ambitious we could build way more interesting where we actually save the engineer the time of even doing the interview um and actually have an AI voice spot that can do the interview so we started with like knowing where there was a pain in our

7:24 personal lives and then iterating from there until we got to something that actually felt like oh this is like a good product which some of it comes from just taste at the beginning yeah before you can like show it to people yeah so interesting so it wasn't this uh Eureka first time out like oh well you know we want to do a startup together and we know exactly what we're going to work on you guys knew you wanted to do a startup

7:44 together you had the pre-existing relationship and now you're out there kind of prototyping searching talking to people doing user Discovery this all occurred before you applied to YC or was this all during this happened maybe a month before we applied to IC very like right right around the same time on my end like I picked my co-founder first and picked the idea second if that sense because I knew I wanted to work a lot of being a Founder

8:05 is like what does the day-to-day look like I think just getting through the day-to- day is like the hard thing for a lot of people and where people fail so just like having someone you want to spend the dayto day with was most important to me and then finding a real pain point and doing the zero to one thing was something I wasn't as worried about I guess the only filter I put on that that someone gave

8:22 me advice on that I thought was very helpful was to find a space where you feel confident you can get like 250 conversations because this person's thesis which I thought was really good I'm trying to remember who it was it was YC founder and I should give them credit but like they were just like if you talk to 250 people like similar people on anything you can probably find a paino and probably figure out like something

8:40 that will have some amount of product Market but just like figure out one is the 250 people that you could actually talk to pretty quickly and like everything is changing right now anyway if you know how to execute you can just like buy the problem so I was lucky in that like engineering managers are group of people that I can like have 250 conversations with very easily as a PM you talk to engineering managers all the time you got a big

8:59 Network your co-founder obviously does as well having come from that background and yeah this is the beauty of Silicon Valley you know people are willing to people are willing to help if you just reach out even on LinkedIn to someone you barely know people often will spend 30 minutes with you and talk to you about unless they're a VC and then they're like it's it's off thesis too early for us to take the cost if it

9:18 comes through my network they're like I got a startup idea and I I already know like oh it's somebody I know I already know this going to probably waste my time but it's going to help them so much oh yeah it's weird to sometimes the world small you know it's so small yeah do the work and then I got to tell my friend like gosh I I don't think I would invest in your startup just so you know

9:36 sorry yeah that is a problem I actually have too even just with business school people because everyone you know know everyone else a startup kind of thing um yeah I'm like dude I don't have money right now I'm I just yeah left my high paying job to do this um what is it like to be in YC with a business degree right cuz you have an NBA and you know YC definitely has this like kind of hacker slant to it

9:58 I mean I think there's two things like I don't know uh two things one is I mean none of us in YC know what we're doing at the beginning right like should be super transparent it's like it's not like I don't know any more or less than any other technical like none of us know's also gone super they've gone a lot earlier over the last like four or five years since I started investing in y they used to take companies that had

10:21 you know 500 K of AR you know they're they're basically like almost on their way to a series a coming they're like a a real Seed Company 500k a million but now it seems like I think Gary tan has kind of shifted a little earlier and he's he's starting to take just basically teams with ideas it kind of goes back to what my company's thesis is which is also what I feel like I'm seeing in the founders it's not really

10:41 about like who knows how to write code anymore or who knows how to do Google ads it's really about who can critically think and like have a good user interview and like get to truth it's it's the same things that you look for in employees is also what a good founder I think would I would like to tug on that thread a little bit cuz I I agree and it's something that I think kind of this generative AI large language model

11:04 hentic movement is is changing about the business landscape which is so exciting yeah it's really exciting and like you don't have to be I don't know very deeply technical to add a ton of value very quickly to to companies and users at those companies you just have to have a little bit of context and like you said be able to critically think and interview them and figure out what their pain points are and then you can very

11:25 quickly get a technical solution together so it's like it's almost like The Disappearance of technological modes we're needing to be deeply technical to get a Hello World site up like in the late '90s or somewhat technical to hack AWS together 15 years ago to get it to work and build something now it's just maybe I don't know what you think about the statement but maybe founder market fit is becoming more important you know like like you said back you know like

11:48 who can you talk to 250 people of in what area and that's a lot of your founder market fit in a way right yeah it I haven't thought enough of about this I thought about this enough for like employees just because we we you know an AI interviewer for employees in terms of what is going to matter more for the future white colar worker kind of um but I imagine it's the same for founder I plug in like every like piece

12:11 of founder material that I have on a company into chat gbt and I'm like analyze the startup and I've done this a few times and it'll come up with this just amazingly comprehensive analysis but it's still like I think to your point what you just said is it doesn't critically think through all the points right it's still up to like a human judgement has to kind of look at everything beautiful mind and make all the connections AI can kind of line up

12:35 all the data for you in a way and and kind of give you like a really good first or second year analyst out of undergrad wood they can kind of line up all the data for you but you kind of need that executive mind to kind of look at everything yeah I think it was the CEO of Jeff blue one spoke T at business and one thing he said um that obviously doesn't matter to me right now is

12:53 startup but it really stuck me as an executive was every decision that he should be making as the CEO of JetBlue is like the 4951 decision and otherwise like it he's not delegating appropriately basically have a D30 decision it's kind of like why am I making this decision if it's an obvious decision yeah somebody on the team should have made it by now those decisions sure it's to make those decisions but there are always going to

13:13 be 50-50 decisions and like are you right like who do you trust in that situation right you want to trust someone who has been specifically assessed to be very good at making 50-50 decisions to the point on like whether like I think for both Founders and for employees being able to make those tough calls really well is what's going to yeah that's really interesting and I think the analogy that I like to use and being a VC is like all the all the

13:35 incoming decks I get to look at right AI can't make a decision whether or not I should meet with each of those companies but it can prioritize do the 7030 or the 8020 hey these that look pretty good you should probably look at these first these ones look okay maybe look at these second and these ones look pretty bad maybe look at these last if you have time and what's the correlator there with light screen right because you're

13:55 you're assessing the critical thinking of the candidate not necessarily the not not necessarily their coding skills but how they think through the logic of solving these these coding challenges I think what we can't tell you or what AI can't tell you today is like will you like working with this person that's like you know will you want to get a beer with this person will this person like fit into your team Dynamic correctly and like be like complimentary

14:18 SKS what we can tell you if this person is technically competent at their job or like at the at the level that you need them to be and like we have faith can like grow into the other pieces of it and I think when I say technical competence do you mean that broadly like in terms of critical thinking skills and ability to explain their code and like things like that like those are all things that we assess forpress going to

14:35 be the right fit for your team in this moment and like even like you as a manager and a employee like is this just like the right for you that's that's not the part that we assess so it's kind of the same thing of what you're saying is I think we could at this point like 40% of candidates is what we're seeing cheat on interviews like an insane number like fact that we just cut out those 40% for

14:51 you is like pretty value ad immediately for companies um and then we can tell you the candidates that can't code or that can code but not to the level you're looking for but then there's this whole set of people that can go to the level you're looking for and at that point you do have to make yourself as a hiring manager it's kind of interesting right it's it's almost like the same funnel as getting into grad school you

15:07 and I both got NBAs so we had to like take the GMAT and all that stuff and you know write good essays and get letters of wreck and all this stuff and if you think about all these like hurdles of getting into a good graduate school or getting a good job they're kind of similar right getting a PM job at Google like pay Consultants thousands of dollars to prep them for that to what degree like if you're an engineer could

15:26 you prep for the light screen screen you know and all thinking to certain correlated with prep and a lot of these excises watching lots of YouTube videos practicing with lots of people and could you get better at this and will light screen in the future help people prepare as well because that's the other side of this this two-sided screening process right one thing that I think is exciting about AI doing evaluations of candidates is that I guess there's two pieces of it

15:54 one is the interviewer because the interviewer is AI can been kind of as long as necessary with a candidate to like really understand what their strengths are and weaknesses are and when we think about what we want lighted to feel like to a candidate it should feel like light spr is like helping you show off your best skills to the company it's not like a antagonistic kind of interview experience it should be like what are you really good at and how do I show

16:16 that to the company whether or not that's good fit for the company is like a different piece of it but what is your magic is where I'd like to get to with light screen and so it's not really about like yes you can prepare for it and all that but really I think hiring is a bit of an art even at like more senior levels so it really is like what is this candidates like vector of skills

16:32 what are the things that are successful at this company in terms of vector of skills and how to much do those things B um and then on your other question around like you know will we go into actually helping candidates prepare for interviews yes but the extension of that is I think light screen can help candidates learn skills that are going to be relevant in the AI future all the things that we've just been talking about can we like help you get better

16:53 and when I learned way back in the day when I worked at Google I worked at Google for Education and one of the things I learned learned there was people teach to the test a little bit like whenever the evaluation metric is is kind of what we end up learning it you manage what you measure sort of thing that's why I started with one of the reasons to start with evaluations is like make sure we're evaluating for the

17:12 things that will matter in the future and then like I think there will naturally become a need also for like re reskilling people and teaching people these skills and I think light screen should also move into them so there's a lot to do because you know if you go through these and I remember back you know 10 years ago when I was interviewing for jobs you they'd have you like do some assignment or some one of these things I I just flat out refuse

17:33 to do them I'm just not going to do it you know like I got I got an NBA from top school I'm smart enough I'm not going to do your work assignment interview me if you if like and see if I'm a culturally cultural fit but I think it's a little different for engineers right because you kind of do need to assess those technical chops right you think the world is changing in a like where you are not just looking

17:52 for us spased companies are not just looking for us spaced Engineers like the honestly the market has just gotten more competitive for but especially hearing I think where people are hiring engineers in Latin America and Asia all over the place where like honestly I think there is like sometimes more of a willingness to do these take home assignments and there's just more of a understanding that like people yeah they're not padas like me living in Silicon Valley and

18:12 they're like I'd be lucky to make the six figure salary at some name brand tech company yeah from a little bit like yeah we don't want to yeah on the flip side of it you know for me I get excited about you know when I was an undergrad I went to pona college which is a fantastic undergrad but it's not like one of those big names you might hear every day some recruiter at Google like found my resume out of a pile and like

18:34 gave me a shot of interviewing and like that only happened because Google has so many resources around interviewing for the APM program like program changed my life and like that only happened because they were willing to like like figure out how to SK interview people at scale yeah actually assess for skills not just for like what's on the resume and I do think resum is I'm become less and less useful over time even at like the top

18:54 level so one of the reasons I'm excited about building an AI interviewer just from you know good for the world is like you know if you can actually scale up the ability to interview candidates and give candidates a shot you're actually building more of a meritocratic organization like at what point in the process do you find your customers using light screen is it right like post application is it like past the first phone screen what do they kind of like

19:16 to insert the screening tool so what we found so far just in terms of stats from light screen so um we've only been launched two months but what I can say is we did one of our customers to do an analysis of the interviews we had done for them and we passed 12% of the the candidates onwards from the light screen interview and of those 12% they were doing us they were using us as a very

19:34 first screen like before they talked to anyone in the company so 12% of the candidates passed that first scen and of those that did pass 50% of them ended up getting offers ratio we just got that data which I'm very excited about early on but like you can see that the signal is really high because I think it's not just coding interview right it is doing like it really is I think because it's able like people are able to communicate

19:53 their things which is just a much higher bar if people are able to communicate their code versus just right there you're only passing one out of 10 and one out of two of those half of those are getting a job offer at least so far that is what I've versus like what was it before they would do how many would pass what was the kind of lift on that I don't actually know what this customer was doing before what I've

20:11 seen on average from what I like historically is you know you might have like a uh each round you only drop like 20 to 30% of candidates you really don't pass only 12 12% you pass like 70% in each round and then slowly get to like your one or one to five candidates that you might give offers so oh dropping it down to like 90% of beginning is is huge uh value and so because of that

20:32 companies are using us very early on in their process they might do a recruiter screen ahead of that like a 10-minute just like check online visas and also like have a face like a relationship with the candidate so the expectations are set but then we're we happen very early because you can use us that scale for as many interviews as you want and then and then invest double the time if you want of your human time on like the

20:50 few candidates that pass right yeah that's that's really awesome what's the rubric look like if I'm a hiring manager now considering somebody that's gone through the light screen process and you know I have the 12% of candidates here you know maybe maybe 100 so I'm looking at the top 10 out of 100 uh how am I prioritizing what kind of reports what's that rubric rubric look like from you guys guess I could try to show you one

21:13 but basically the rubric has three you can see this now so basically what you're seeing is what the evaluation looks like that a company gets a few minutes after like scen interview is complete normally there's a full video of the interview in the top right corner you can also here see here you can see the candidates code over time you can see the transcript of the candidate and then our AI who's named Ethan who like will provide head sometimes or you know

21:35 we'll do kind of what a traditional human interviewer would do ask about time complexity and followup questions and make things harder depending on how the candidate is doing and then we have three buckets here that is behind this is like literally hundreds of questions that we are asking the AI in order to pull out these buckets but you know what what's happening on the front end for for the hiring manager is they get an overall score on the candidate score out

21:54 of one out of four and then they get the candidates kind of strengths and weaknesses in a short summary at the top wow why' you guys do a a fourpoint scale instead of like five or seven Gavin and I both come from Google so a lot of this is heavily inspired by Google's rcks to be completely honest Google does I believe the reason Google does a four-point scale is because they really don't want to they want they want a

22:15 bunch of Threes exactly an interviewer to make a decision one way it's good or bad it's like above average or below average you can't have something you cannot just like basically yeah yeah we also don't allow ourselves to we want to like make sure we're providing value to company we do the same thing when we're evaluating startups actually for the same reason we have the same thing even if the cheat detection actually because people are cheat in all kinds of

22:35 interesting ways now we we used to have a signal where we would have like a yellow flag where it's like oh maybe they cheated and we're like this is dumb like either they cheated or they didn't we have to stand by our yeah decision we say one through fives no no three no threes basically so effectively it's a four point scale so you have to do a two and a half or a three and a half you

22:51 cannot do a three I see you're allowed halfs though you could do halves yeah we could do hals with our scales effectively it's like a 10o scale but anyway cool uh we have three categories right now overall which is problem solving technical skills and communication collaboration the technical skills are I think more stereotypically what an engineering interview would have which is just are they using the right data structures do they seem to know how to code but I can

23:12 we stop and let's just appreciate how advanced AI has gotten in just just a few years here that you can build something like this in a few months effectively right you've been launched for a couple months YC was what three or four months long and you started working on the idea pretty much before you got into YC and now you have this fully built AI interviewer that can assess people at scale for their technical capabilities return a bunch of natural

23:35 language and ratings areas that they're good at areas they're not so good at I mean it's pretty incredible first of all I'll say shout out to Gavin Gavin is really an excellent engineer and that is why this is so good I don't think it would be so good basically most other people made this but the other thing that I'll mention is you know Sonet and some of these models they've been trained on so much coding data in

23:54 particular that they're really good like they're actually exceptionally good at understanding code in real time CU in real time this interviewer is like seeing where you are in the code and like deciding whether or not to provide you hints or ask you follow-up questions or decide it's not scripted right it's just just based on the question and like there's a recursive Loop running where it's evaluating every what multiple times a minute about like what's the status of this interview right should I

24:17 say something should I not like we have a model specifically for that to make sure just for the should I speak or should I not speak and we have to do that because we want to make sure the candidate experience is so good W there's there's one model just evaluating whether or not I should speak to they that's amazing so it's it's a multi-agent model but that works yeah I mean effective yeah effectively it's that AI can do this in like sub like the

24:41 latency has gone down so much and I think as voice to voice gets better and better we'll be able to do more different kinds of interviews you can imagine instead of having a coding interface a spreadsheet interface a whiteboard or just like open use your monitor and use your own tools and we'll just like watch your SC yeah there's a lot of things that we can we can do and yeah you can just analyze so much about

24:59 a candidate now that you couldn't before yeah that's pretty wild yeah and the latency like even two years ago when chat gpt3 came out you couldn't you couldn't do this right and now yeah you couldn't do this even like a few yeah like even in June of last year I think you couldn't probably do this yeah the latency would have been so slow and yeah laty would have been slow the voice quality wasn't as good um yeah I coming

25:20 in token cost of running these models would have been very too expensive and totally voice generation used to be really slow and voice generation has just gotten so much faster which is also a huge piece of this so we've been able to you know our turnaround times are like less than that's amazing so about a month you started wrenching ideas on ideas before a month before you got into YC at what point you were like oh what

25:39 was that aha moment was it interview 250th was it the 50th interview how many of these interviews did you go through you're like this is it this is the problem when we right before we launched on YC the time it became like oh this is it was like we launched on YC's like LinkedIn we got literally 40 meetings booked within a week it was like insane the number of meeting we had both that week and this this is their launch feed

26:01 their launch feed yeah they do on LinkedIn within like especially that first week of meetings where I think the most high- inent people like we would show up to the meeting we would just do the demo that was already described in the LinkedIn post we were just I think proving to them that this product really existed and people would commit like that like in the meeting in the 30- minute call that they were ready to give

26:19 it a shot and I think that's when we were like oh this is something because it had been so hard I think this is like advice that I've heard from other people and I'll just reshare it because I think it was really help us is it if you're like pulling teeth to get customers to use your product you're probably not where could fit but if I've been in this place I'm like oh no it's because I'm

26:36 not following up correctly my messaging is bad like I'm not using the right channels but like if people aren't coming it's probably they don't want it versus in this case finally we had figured out something where we were like oh people want to like people are ready like today to start using this can can I use it now like how much does it cost right yeah so questions like 15 minutes into the first call is when I was like

26:58 oh this is like a real thing this is a hair and fire problem that people want they have a budget to solve it right now there's a real cost you know engineering costs are among the highest in Tech you know hundreds of dollars per hour let's say in Silicon Valley and I really think feel like cheating and the the fact that skills are changing for interviews is like especially a hair and fire problem for people yeah tell me tell us more

27:18 about the cheating because that's the second time you've mentioned it it seems to be something yes it seems to be and I remember when we originally spoke you talked about it and both of these are really big problems that I I think like people have been saying that engineering interviews are broken for like kind of since I've been in the Tex phas like for 10 years at least people are like Elite code is stupid whatever but like that's

27:36 true but like the fact right now that like 40% of candidates are using chat gbt on their on their take-home assignments or on their even in their inperson interviews well yeah I mean you could just download anthropics like you know screen agent it could read the screen and just give you advice on the side there's even specialized tools now for for candidates to buy that are really expensive to just like yeah exactly it's a voice agent that's

27:56 listening on your interview and tell you exactly what to say kind of thing or like you can just copy your prompt into chbt and change the output a little bit I remember doing this on my own interviews like back when I was doing PM interviews I was like I don't know how many golf balls fit into a 747 you know yeah more remote interviews right where it's easy people to be typing at the same time and like I think what that

28:15 really is a signal of is that the interviews are broken right we're we're testing for skills that are not actually the skills that are going to matter people can like find that answer in like 5 Seconds using the internet so it's an opportunity for us to change the interview format or change at least what we're evaluating the interview to something that's like like less treatable basically and how does the cheating detection work is it another model that's that's somehow detecting

28:36 and do you prompt like as a reminder please don't use internet sources like what what does that whole process look like sure so I'll be a little vague about this because I also don't want candidates to be gaming our cheating detection but there's three buckets of how we do our cheating detection today the first bucket is kind of what I I call like the Via signals I guess it's like if someone copy pasted a bunch of

28:56 code we can detect that if their Mouse you know we're tracking Mouse position tab position cursor position or Focus cursor position and those are all signals that weuse you can tell that they're all tabbing into different Windows tabs and Y totally yeah and then the second set of signals which I think is actually the most common thing that Flats candidates is like what I call statistical signal so this is how similar is this code to what comes out

29:19 of the models like our reference Solutions basically and then also what other candidates write if the code and then also with that in the statistical analysis it's it was the code incomplete top to bottom left to right with no going back lucky for us code is often not written linearly if people are really writing the code right they go back and forth a little bit but when people are cheating they often go like top to bottom completely and so those

29:39 kinds of signals basically on like how the code is written how similar the code is Solutions is like bucket two and bucket three is on the conversation side um and contextual kind of signal so the fact that the AI is asking you questions and so you know if you answered one question really fast because the AI told you the answer and then the AI was can you explain what's happening on line 12 and it takes you a long time to figure

29:58 out what's happening on live 12 that's the sort of signal that we use as well so it's pretty very hard for people to cheat on that's incredible so and so the candid doesn't know that you're detecting this you just tell the your your customer so we don't tell them in the interview that you're cheating we do tell them before they start the interview we have a huge thing that says please don't cheat on this interview we

30:16 really will catch you like we're industry leading people still try to cheat it but uh yeah we don't we don't we don't call them out during the interview we just tell the so how does how does this approach differ from other tools like there's hacker rank there's ATS systems you mentioned another one and you know were they just contributing to the interview process being broken like how do you how do you think about your differentiation there's

30:38 a few things here I think none of these other tools that are out there are really trying to be a drop in replacement for a human interviewer as far as I can tell I think they're really looking at their tool as like a a whittling down tool like okay you put in thousands of candidates you get out hundreds of candidates and now you do the full interview process the hundreds of candidates so it's just like meant to

30:55 be a way for you to like make sense of the noise is what we are really trying to do is be a replacement for the human that's actually doing the interview that's assessing a variety of skills not just the code quality and so what that means is companies can send us way more complicated questions than what you can do on hacker like you actually can do the sort of questions that a human would do in an interview might be like rest

31:13 API or like whiteboarding kind of solution we're evaluating of course a lot more things and we're we're trying to replace the time of a human interviewer versus uh like kind of being a sass product that just has a large question bank that like companies can just candidates through that makes sense yeah so that's that's I think the main differentiator and I think for us like what I think about longterm for light screen is we want to be the trusted

31:34 source of Truth on candidate skills so you can imagine a world where we actually credential candidates for certain skills and the candidates are able to go instead of their resume or in addition to their resume they're like list of Life screen credentials and companies actually trust that because we actually verify the skills we've built up that we like understand how to detect candidate uh like detect shooting and also like evaluate skills the companies really trust the evaluations that come

31:55 out of us the way they like have trusted historically like you know High at institutions and this is something that nobody's really I guess hacker rank tried to do this a little bit but you know it's kind of like that G2 crowd or glass door but for candidates right and I guess LinkedIn tried to do it with some of their LinkedIn learning stuff some of their yeah there's like SK kind endorsements and stuff but yeah or

32:15 Udacity or Corsair with certificates and I it's technology didn't exist before like I don't even blame them I think they they just didn't have a way to have an AI that's actually going to see what's happening on your entire screen evaluate this level this number of signals on candidates and be able to actually do it back and forth rather than just being like a static interface so I think they attack before to do what I think people have known for a long

32:37 time would be valuable it's just like we couldn't do it even now we're kind of on the edge of I think you know whether or not we can do it betting that 03 and some of these other models we're going to make this better and for us oh yeah yeah that's wild anything that you would you know you would tell somebody starting YC or thinking about applying to YC I get I get this all the time

32:56 because we we have YC Focus fun we have like 60 YC companies so I'll get people reaching out for my network like Hey how do I get in a YC or like is it worth it or you know in hindsight what would you tell somebody I have a lot of advice but I guess you're not too early just apply you worst case you'll get some useful feedback hopefully out of it and it will be a good exercise for you anyway to do

33:15 the application process so I think we were concerned we were a little too early and you know like YY is a great place for you to like figure out if your idea is good or not the earlier you do the less time you kind of waste so I'd encourage people to just apply and then in terms of is it worth it I mean it's even for me and Gavin where I I kind of thought maybe it wouldn't be as worth it

33:31 for us just because we have been in the industry for a while and honestly do have some of our own connections already we it's so helpful to just have like smart people who've done this before telling you the obvious stuff like it's not like they're telling you some rocket science it's just like focus on the customer make sure you're getting them to pay you like simple things like that but just having someone say that to you

33:50 really deliberately that you respect it does really push you push you forward and the other Founders being there to push you forward and like you know kind of friendly compete with is do they have like an internal kind of leaderboard so you know kind of who's doing the best in in the batch or is it kind of like informal they don't really have that but they do have you know we have every two weeks meetings with like a pod of like

34:10 15 or so startups and you kind of hear how everyone's doing in that pod so you can kind of get a acely try to and I think you you know it's a long game versus a short game so yeah you try not to compare yourself in the moment but I think all the founders are naturally competitive we can't help ourselves right yeah which is good I think it just pushes to be and I like what you said there too it's it's really

34:29 a marathon it's not a a 100 meter dash this this is like a 10year journey any startup that you're going to do or longer and you know I've seen a lot of companies flame out you know they're they're kicking ass coming out of YC they close the around in two days and they got a million of AR and then they just like like go sideways for for years yeah and I've seen the the opposite happen where they're just like

34:47 floundering for a year or two just trying everything trying everything grinding grinding grinding and then all of a sudden they take off and everything in between there's um oddly let is a very strange business school where I feel like I didn't learn a lot of business but I learned a lot of just how to keep your mind right in difficult situations and um that being such a consistent art uh sales and also of leadership of just when things are good

35:12 enjoy it but also like don't go crazy but when things are bad also don't go crazy and know we're working in hiring software so I think it's particularly attune to like the ups and downs of the market a good time I think I think you might look back at this and be like we we nailed this right at the right time because that's something I think about as an investor like because hiring was so dead in 23 24 and now you have this new

35:35 technology that unlocks a new way of doing it in more efficient ways saves time saves money and now all of a sudden hiring I think it's going to come back in 25 26 and so like I think you'll look back and be like Oh 24 was like a really good time to found this company that is part of our hypothesis too yeah love that well we're we're in cahoots there why don't we wrap up with the rapid fire

35:55 sure let's do it what is the most exciting application of AI you've seen outside of what you guys are doing I mean some of the stuff you know you've seen like the AI advertisements that are out there like yeah where they it's like Minority Report and they customize it to you as a person yeah Custom Custom imagery custom videos hey Brian picture your family like on a you know Mediterranean Cruise what how do you know like yeah morally and like even

36:19 like you know you can like have better representation and like speak to the specific person's problems I feel like there's actually a lot of very cool stuff happening and advertising it in concentration what book or resources had the biggest impact on your entrepreneurial Journey okay so books that um have been really helpful for me on my entrepreneurial journey I really like biographies of entrepreneurs so I read Walter isaacson's book on Elon recently genuinely fantastic biography good I read his on CH dos I read this

36:48 stuff on Jennifer DNA like all of those have been great even though Jennifer DNA you might not like think of as like a Founder but really like a leader in her field is one of the crisper innovators and it's cool you get to learn a lot about like their space and also you get to see someone's like leadership challenge really up close amazing yeah I love that biography as well what's one underrated skill that candidates should

37:06 focus on showcasing and their technical interviews communication skills I mean this is and it's not what I mean by communication is not seem likable and do small talk it's like can you explain your code and can you explain how you come to this solution youve come to just like talking out loud more it's amazing how many candidates s to this day are like okay give me a minute to think think for five minutes and then just

37:26 like write perfect code like that is not going to help you in this interview and it's not going to help you in life you have to bring people along with you on the journey and I would really encourage candidates to to just talk more as they're thinking that's a that's an interview hack actually that I used to deploy is instead of just silently thinking about it I just be like oh okay you know and just think out loud and I'm

37:44 a think out loud person anyway worked out for me but yeah I think for sometimes it's not natural push your know besides light screen as a product leader what's your favorite tool you've ever worked on you're Mak make ches between all my children I mean working at a assembly AI was really exceptional so at assembly I was the head of product and assembly is a API for speech text and speech understanding products and what was awesome about that is that I

38:06 love working on API products not I'm not doing it right now but I enjoy it because it's a building block for other people to build their companies so you could just see what all the voice spot companies were doing with audio data and it's so cool you got to talk to remember the first time we talked to Spotify Spotify is one of assembly's biggest customers and like you're using us for some like ad placement algorithms and it

38:23 was just like wow they really need this like this not work with out assembly and like when you realize just like how much they need the product that you're building it's very inspiring I think assembly is probably one of the coolest things that worked done but honestly I mean r OSS was so cool gopuff was like everything that I've got I've been very lucky with the sort of things work yeah how do you how do you balance as you

38:42 think about kind of the next chapter of scaling maintaining a leanness to the startup being a Lean Startup versus like scaling up the head count like how do you kind of think about this next chapter I think because I come from being in the valley over the last 10 years where been like a bit like a lot of ups and downs um during Co and stuff where I'm really focused on staying as lean as possible where it's not

39:06 inhibiting your growth is kind of my framework for it it's like if I'm at a point where I know one more person would actually mean one more like like specific Revenue I'm like okay I'm GNA hire that person immediately and I think what that means is you're kind of always warm hiring like I'm maintaining relationships with my network and like if someone amazing comes along of course I like figure it out but like in terms of actively doing hiring I am doing it

39:28 when I know that like there's like a set of work that we could be doing that' be so impactful and we're already losing out and actually that's honestly like we're pretty close to right that right now on the engineering side so I think we will be hiring an engineer pretty soon anyone anyone great let me know yeah yeah of course I'll help what's a misconception about AI in hiring that you'd like to debunk I think some people

39:46 think jobs are just over I think people think AI is going to take the engineering jobs or take all of the jobs um or the white collar jobs and I think I mean that's like a cool Utopia maybe but I I think that jobs are just getting more and more abstract and or they're change really dramatically I think people it's not like people like to work but it does it gives people a sense of purpose and

40:08 it does like structure Society in a way and also it is to the conversation we were having earlier about the skills that are going to be necessary there are things that you do really trust a person for that you don't trust an AI for and so I or at least a human with an AI you know with an AI and I do it I do it all the time as a solo VC the solo GP VC

40:25 like I'm talking to the AI about stuff totally you know I'll be on a call with a founder and I'll be like tell me about this space you know and I'll be listening to the founder and I'll be kind of checking it with the AI what they're saying and it become yeah it becomes like this companion co-pilot if you use the Microsoft term but you know it does it's it's this thing that I get to collaborate with and work is changing

40:46 but I think to your point I don't think it goes away either I think if anything there's a jeans's paradox you know where you know economically the cost of inputs are dropping therefore the quantity ad maned will go up and you'll you'll see a supply of software you know infiltrating every nook and cranny in the market and yeah it's an exciting time to be an investor because of that with less Capital needs um so where can folks interested in

41:10 using light screen or or just generally uh to learn more I find out more yeah if you go to light screen. a uh there's a ton of information on our website and then you can also book time with me directly there and I'm happy to show you a more in-depth demo of of what we've built and how it could be helpful to you well this was an amazing conversation thanks forh so much for coming on surey

41:28 to be here thanks for having me Brian yeah thank you