He Built an AI Job Search Agent After Getting Laid Off — Now It Has 30K GitHub Stars

0:00This guy got laid off after having a baby. He started job hunting the usual way, and then he came up with a clever way to use Claude to get a job. His system scrapes job sites, finds the jobs that are the right fit, customizes everything like cover letters, and automates much of the job search. It actually helped him get a great job. He then did something incredible. He posted [snorts]

0:22his setup on GitHub. Thousands of people used his setup to get jobs. It became the top GitHub repo of the week twice. Now, we're going to do a screen share where he'll show you how this system automates a job hunt and how you can use it, too. But, the other takeaway that I want you to get from this is that the new way to share and teach and [music] help the world is to

0:44just publish your workflow on GitHub. You're going to see the impact it has. Let's get to it. Presented by Zapier, the AI automation company. Mads, what are we looking at over here? So, we are looking at the job search dashboard, uh which actually shows you the current state of the different jobs that you have actively been applying to. So, you can see the ones that are active, interviews, offered, if you are hired, or the ones that are rejected and closed. And uh yeah, the different sectors that you've been applying to by channel, uh and you have all the applications

1:17there and also a nice tabular form. It is so cool, and one of the coolest parts about this is that you didn't make this. Who added this? Uh well, because it's open source and it's MIT license, uh there's a lot of people that try to to come up with contributions and and really creative ideas of how to represent all the stuff

1:36that is inside the the repository. Um so, this is just one one of many contributions um lately. Yeah. So, I love this. This is so exciting. Okay, let's take a look at that second tab, which is LinkedIn. What are we looking at with that one? Yeah, so for this one, it's uh one of the um applications that was found using this job scraping workflow that is also part of the repository, right? So, initially you you set up your profile. There are different ways to do that, but basically you can you can one of the approaches is to to take your former CVs and cover letters and all that, and then it would

2:12nicely sort it out what is going on and build like a bulk profile of who you are as a person. And based on that profile, you can then proceed with finding different suitable roles for you. And this is just one of the examples of an So, I tried to make like a demo profile to potential role for me, and this is one of them from from LinkedIn, a machine

2:35learning engineer role. And the way that you got it by was by running a prompt in Claude. You want to show that and show the output? Yeah. Sure. So, the prompt that I just wrote was this job scraper. And when when you initialize that and you have your profile and you have figured out different sectors or whatever different roles that you want to apply to, hitting this command will then go through the different portals that you've set up, for example LinkedIn, and come up with with different potential roles for you. So, this is just the first one with some of the higher

3:09matches you can see here. The different titles, the companies, the locations, deadlines, the link to the URLs where the the scraper has actually found them, and you can see the medium matches and some of the Usually there's also the lower fits, right? But but you should you would you would pay your attention to the high

3:25matches. Yeah. Okay. And what it's doing is it's scraping individual sites. I've asked you in the past which ones, and you've explained to me, "Look, different countries have different ones that are worthwhile." And so, you we let the user pick it. All of this is in a a GitHub repo that anyone can see. Would you just kind of scroll through it and just show us just a quick overview of what's in it, and then we'll go into the next step. Once you've scraped a job and you've found it, what's Yeah, what's in

3:49here if you want to just show us? In the uh, itself? In the Yeah, in the repository. It's like set of skills, right? Yeah, exactly. Exactly. It's a It's a set of skills and there is a lot of different commands and, yeah, this is just some of the stuff that I just, uh, covered right with the setup. You fill

4:03in your profile. You do the scraping. Uh, presents the the matches. You could do a fit. We'll come back to that shortly with the, uh, slash apply, which is really like the I think the the holy grail of, uh, of the repo. And then, um, after that, you there is a um, draft and review agent that that helps you like come up with the best version of software for a specific role that you

4:27want to apply for. Okay. All right, let's take a look at the next step in the process. And again, just to say, all I usually do is I take the URL from this repo. I give it to Claude and I say, "Okay, I want to do this." Okay. So, now it scraped the job site. It found some good, uh, prospects for you based on your background and now you pick the one that you think is good and what does it do

4:49next? Well, then it, uh, when you use the slash apply, um, command, it then fetches the the job description or you can you can fill it in yourself. Um, or if it's a part of the conversation where you've just scraped the the the different roles, you can you can just, uh, continue from there. But as you can see here, it says fetch this, uh, LinkedIn, um, job description that I had before. And then it comes up with the evaluation for for this specific role. And there are different dimensions that it actually tries to match you on, right? There's the technical skills, the experience match, behavioral fit, how

5:24you are as a or, yeah, as you are as a person. Uh, do you like bigger teams, smaller teams? What kind of work environment? Uh, the location itself, is it uh, within your commute constraint, for example, if you if you put in that. Uh, also your career alignment. And you can see for for this specific role, it gives me an overall score of, uh, 73 out

5:44of 100 and says that it's a good fit. And then it with highlighting the key strengths for this role, what is some of the advantages perhaps that I have, and it also has the gaps that you want to address, right? I mean, the the things where perhaps you you fall a bit short, but it could also be how you can you can comprehend with this or how it should be approached, for example, in the cover

6:05letter. Uh so, the overall recommendation here is that I should apply. And it's a yes with some caveats, right? So. I like this a lot. Okay. So, then what I like that it that it does this type of analysis and it does it in bulk. Yeah. What's the next step once you do this?

6:26When you say apply? Yeah, if I then you see apply, then we come back to to the holy grail of the the draft on the reviewer agent. So, it it it comes up with the the drafter that yeah, drafts a version of your cover letter and your CV for for this specific role. And then it's sends that off to a reviewer agent which has a fresh context window, so no no information of who you are and it's it's not really biased in that way. And it it basically just takes the the parts from from the draft agent and really scrutinizes against your ground profile. Is this really true? Is

7:04this something that Mass has done? The the things that he's claiming? Is is there something that is perhaps blown up a bit for this role? So, it it tries to make it as grounded as possible, but also, of course, being again presented in in a nice manner, right? Uh and then then when the reviewer has come up with this critique, it's turned back to the the drafter and yeah, the there is a finalized version

7:28presented to you in the end. Can we see the the CV and the cover letter? I've got a couple of questions about that. Yes, let's see. Because the beauty of this, I've tried this myself, it will customize to the job. Yeah, exactly. So, it it really tries to to highlight all the things that you have done that is relevant for the role, right? So, it's it's a it's a process of um maybe not just highlighting your your latest role with the perhaps the the most relevant things that that you have done, but also if there are former roles that are also relevant, it can also cherry-pick all these things that are

8:03that are relevant for for the the current job description that you're looking at, right? Uh so, yeah, selected projects, projects, education, some publications for my side here, certifications, professional experience, and all all these things are are presented and are cherry-picked for this specific role. I found that the CV was especially well done and I didn't even have to say anything beyond here's the job. Um okay, and then who actually clicks the buttons and applies on LinkedIn or the other

8:34sites? Uh that's I mean, that's that's my philosophy that it's not something that I've implemented that you you also apply. Uh this is something that is presented to you and you are the the owner, of course. You you are at least the the one that wants to be employed. So, uh in the end, it's it's up to you whether you want to send this or not. So, that's why I've actually set the threshold that it's up until here that we made the optimization process, and then from here you you need to apply

9:02yourself. Um and so, the dashboard was created for you by someone, someone who contributed in. If someone were to contribute the the auto apply, would you add that on or would you push it away? I I haven't added it on so far. I cannot actually remember, I mean, there is over 200 pull requests by now. Um I I would say that in in my main philosophy and also what I wrote about while I was unemployed at LinkedIn was that this is I still regard it as a a human process in the end, Uh, although there's a lot of this stuff that is

9:37filled in with AI now. So I like the idea of still having some kind of human in the loop just at at some point in in in the workflow, right? And that's also why I came up with this evaluation fit, right? I mean this is not a by volume game. I I myself applied for I think it was 69 different roles and I was invited to around 21st interviews and I think I mean a conversion rate of of 30% or so roughly is perhaps pretty good. But yeah, and of course it could also be about my

10:14profile. But but I do think that there is a lot of how can I say a lot of the the process was also filtering out the unrelevant roles for me. So this is really about yeah, looking at this evaluation metric. Doesn't really I mean the the specific number whether it's 72 or 73% is not really relevant. It's more about making yourself think about what you want, what

10:38is good for you. I mean what what what kind of job do you actually want? So I really like the idea of having an AI that also pushes back and not just mass applies for you. So then how does it know in the dashboard how many jobs I applied to and what I'm waiting on? Well, it is because it it keeps up some statistics in the end about the the different roles that you have applied

11:01to. I mean this this was I applied to? Yeah, because I mean everything is in your contact window of course. So when when you do the apply you can then set say to it afterwards, okay I actually applied for this job. Please log it in our memory or something. I I think there are different

11:18ways to do it. As I said this this dashboard was not something that was uh um something yeah, exactly. It was contributed in. But I mean I I I personally myself just kept a CSV file with all the logging of of the different jobs that I had applied to and the ones that I was rejected from and all that. So, yeah, just trying to keep like a small diary so to say of of uh

11:43the job hunt. So, the interesting thing is you put it up, it became the number one GitHub repository of the week for 2 weeks. Yeah. You didn't know about it until we covered it here on the channel and that's when people told you. The reason I'm shocked by that is I thought what you were doing was putting it out there to kind of raise your profile or at the very least putting it out there the way some people post on LinkedIn or Twitter to let you

12:09know, to create content for the world. And then we watch [clears throat] to see how many people are viewing it, how many people are liking it and so on. You didn't do any of that. What was your motivation for publishing and why didn't you pay attention to those vanity metrics? I think I've been so happy about the open source community myself and all the things that that people put out there on on for example GitHub with the repositories and how they they share the knowledge. Um so, I wanted to give something back and well, it wasn't at the time when I published this repository. I thought

12:43like, okay, this is going to blow up and have I think it's about 900 oh sorry, 9,000 forks by now and we're approaching 30,000 stars or so. But but I mean, just having one or two persons finding this stuff useful or taking it with this MIT license and using it for for their own workflows would at least make me happy and the the way that people have reached out to me from all over the world basically saying how happy they are for this open source workflow is just yeah, overwhelming. So, it's it's really something that gives

13:23back I don't know, a million fold or so. It's it's really amazing. So, this was just you saying, "I've gotten so much from the open source community. There's times when I needed some skill and I would go and use it. There's times when I needed an app and I would go and get it." And so, now I'm just going to contribute it. And I did see actually you had a couple of other projects that you put up on GitHub all public none got nowhere anywhere near this. [laughter] We're talking about

13:44dozens of stars versus now thousands. Yes. Um okay. I think that this is the future of teaching people. In the past, if somebody learned something they might write a blog In the old days it would be write a book about it, right? That's how you share knowledge. Yeah. Then it became further democratized by

14:01you could write a blog post about it. Then it became even bigger by creating a YouTube post or even an Instagram post. And it was a way for people to teach and and others to learn and at the same time raise their profiles. Yeah. What I like about this is you are not just teaching, you're giving people the tool that gets the result that you got. You're basically you From what I understand, you went through Claude and you said do all this. You kept changing it. And every time you just come up with a new improvement, you improve your personal Claude skills. And at the end you ended up with skills that

14:31made sense. And you said, "I'm not going to use it for myself. I'll package it and I'll give it to other people. They get the same results as me instead of writing a blog post saying what I did." That's the vision, right? It's also funny if you actually look at the Now, I don't have the statistics about how much of my repositories code and how much of it is actually just plain text. I think it's about 50/50 or so. I mean, it's it's funny that skills I mean, it's it's just built up by words, right? I mean, everyone can can read it and and understand it. So, it's

14:59it's um it's a it's a new world where these highly specialized instructions for for the AI models is something that is shareable and yeah, as you said yourself, the the new way of sharing knowledge, I think. Or at least filtering out or dictating which way the the AI model should go. Um what one of the things that people also uh reach out to me uh asking is how the how you how you set it up basically, right? And I think there's also a lower threshold now of or the barrier is is lower now to uh how to make these stuff uh or how how to make it run on your own

15:41computer. Because basically you you can just take the URL, the the URL for the GitHub repository, paste it in your your prompt and say how how do you make this work on your local machine and it will guide you and it will just set it up for you, right? So. All right, let's close out with this. Can you just tell me your story of how you lost your job, what you did, and how

16:02this helped you, and where you are now? Yeah, exactly. I can do that. I am I was impacted by layoffs in in mid-December last year. And at the time I was yeah, actually on parental leave and I was uh I was playing [snorts] around with with Claude Code because I saw that I mean a lot of things were were going on with agentic

16:26coding at the time. And um after the the unemployment started, I also talked to a career advisor and tried to figure out what I wanted to do and I mean all this combined basically made me realize, okay, this is actually something that can be optimized and this is something that uh agentic workflows can can help you with. Uh so it it was also like a trial and error process because uh in the beginning I was just having the the draft uh agent and I saw that, okay, the the outcome is actually not that good and it would just blow up all the things that I've done. I was like,

17:06wait, I haven't published that many articles and stuff like that, right? So um I think the the really strong idea came when I was looking at how, for example, um RAG systems and and different stuff uh uh you you use LLM as a judge. So, you you can use another agent to um critique or score one agent's output. So, using this basic idea was something that I then tried to implement and I just saw that the results became way better. And I thought like, okay, this is at now it's something that I am not ashamed of making public. Uh so, if any anybody else can use it, then

17:45then yeah, feel free. And that was while I was job hunting myself. So, uh yeah, then after a couple of months of using my own workflow and also trying to tinker with it a bit more, I managed to to land a job as an AI engineer. So, uh [laughter] Yeah. and that's where you are right now.

18:05Yes. I wanted to highlight this partially because I think this is a really good repo that has helped thousands of people, and the other part because I think this is the future. We built things for ourselves, we turn them into skills so that they're reusable, or our process into skills so it's reusable. And now the next step after

18:22that is to share it, and you've done it. It's been exciting to see how the community now is building on it in addition to using it, and I should say to anyone listening, this is one of 10 repos that we covered the first week that that the repo became top 10. We do this every week where we go through the top 10 repos of the week, and we see what's hot, what's interesting, how you can use it, and there's so many great gems in there. And so, I'll have a link right here so you can see one of them, and I think you'll well, I'll see you in

18:49the next in that one.