Summary
Claude Desktop Project Setup:
Prompt Template Shared:
Key Prompting Insights:
Claude Skills:
LaTeX Workflow:
IDE-Based Workflow:
Setup:
ATS Scoring System:
Results:
Job Tracking Web App in Development:
Live Demo of Agent Creation:
How Deep to Go:
Technical Stack:
Features:
Notes
Transcript
Thank you. Basic presentation, lecture capture, what is this? Wow, okay. For today We have a littleIt's a mix. This is a experimental session. But yeah, no, no. Let's see.
So this should be taking notes as we participate. And then at the end, like we are sharing those notes with everyone who could not make it. And also I'm gonna start recording this session on Zoom. Great. So we're going to start recording this also. And This is my screen share on Zoom, so I guess It's working there. So Just for now, I want to give some updates of what this group is about. So we're getting people on board of creating some, like making sense of what is AI agents in AI and how can we use them for our own purposes and also like in a good way and a way that we can make the most of it while we can. So for most of us, this is our last quarter, so we want to get some experience on it. We just had one session before that was like to getting to know what questions we had about AI.
And I think from that session, we have this session note. So I think this could be a good way, this page here, a good way to maybe document our journey and to share like every session for everyone. We'll show that with you and maybe get some feedback or comments if you want. So here you will find like the join WhatsApp community. So in the community, for those that are on Zoom also, or whoever is watching this video, hopefully at the end, after this session, if they couldn't make it, you can join the community. And on WhatsApp, we're going to share like the sessions that we're going to have.
We're aiming for one session each week. It won't be like to like too complicated or it should be light and also for us to share things that we think are interesting and we want to learn more about agents. So you can find that here. And there's like... Currently there's an availability poll, so we're finding that sometimes people don't have the time to join us on Fridays, so you can just basically here, only for people who have their ability, but here in the page you will find this time slot to fill, to be filled like here.
So on top of this community or the idea of this community, I was experimenting using this Obsidian thing so right now the idea is this could be like the brain of our group and how this could work is that right now this is like the nodes are the resources for the sessions, but also this will be fed by the scripts of the current sessions. So the idea is for you to all be able to access these notes and we have this like a common resource to look up for getting to, yeah, for using this thing.
So for example, here we have sessions and this second session here is session two. Agents for Job Applications. There are some goals, some setups that you can have. But I think this could be a good way to to do the sessions and if you have comments or feedback on that that's also super welcome. And the idea is for this first session to show and tell how are you using agents for job applications. Yeah.
How to start? That's a great idea. So the first session we've been Basically, just like worded out, like, okay, what does everyone know about AI and what do you not know but are interested in knowing? But we didn't get a sense of like, I guess. We didn't get a formal sense of how many people are paying for subscriptions or what like a tally of what current tools people are using so I think Yeah, I like that idea. Especially to get everyone started, we can just get a sense of where...
So yeah, can-We'll lead a session on his walkthrough. But I think before that, we could send a poll maybe three or four days in the group chat and just see what-That's good.
Honestly, right now, I use an agent for helping me create the structure of the sessions or the knowledge that we're going to have. And I think the initial server, I think, that came out was fun because now it can explain, like, okay, what are cloud skills, for example? And we've maybe heard of that before if you're using cloud. And it's not entirely clear how to use them or what is it for. This today application will be useful for us to understand how to use them. So yeah, in summary, these will be some knowledge cards. And if we have these built as our brain, that's what I was referring to, this will help us make connections between the things that we're trying to understand and the things that are available out there, right? So this is gonna be that.
point for us. So for example, today's session, We are going to present some of the context. So for job applications, how you can universally use these setups or these workflows and For example, a live demo on me, how I'm using it. also shared with me this week that he was using it in a way that I found really interesting. And he will join us at some point. So we'll have three settings, for example, one for resume, plus cover letters.
Then we'll have a second setting of job search matching and interview preparation. And third setting on end-to-end pipeline, right? So from the start to the end. There are a lot of things that could be new for anyone. And so if someone is on the Zoom call, they also can participate and raise hand if you have questions about any of these topics. But the idea, the second part of the session is to build it your own, right?
So even if you don't, have your own set up ready at this point, we can just figure out how to start or doing that for your job searching, right? So we will build that at the end of the day. And here we will find some resources, for example, about cloud skills, best practices, So how to be specific, how to use clear sections or even iterate like for your drops. And that will be here, but also if you find that some of the resources here are still missing, please tell us and we can explore it together on how to use them.
Because I think So one thing that is important to clarify is we're not telling you like you should subscribe to cloud or to course or whatever solution are you finding, but the idea is to present like different setups here that you may find interesting. And if you find that one was super useful, then you may want to decide on investing on that. But we don't want to like to you to definitely have to pay something.
So, and we actually are exploring options free, like free options of how can you deploy your own servers and use those for your free version of LLMs to help you. Some is by an open club, but also other different free models that can be out there and how can they help us without having to commit to pay to like a monthly subscription. We will do that probably at the end of the quarter because it needs, it requires to set up your own server.
But at this point, so what do I normally use for job creation is to create a project here and here you can have all the projects but So the idea can be born-Yeah. This is the pro version, the other, like one you pay for a subscription. Just, yeah. No, this is cloud desktop. Cloud desktop, yeah. So I download cloud desktop and then start doing like a project, right? For a new project, for example, you want or you may want to create a new project for your job application.
And then in the project, once you create like here a new project, you can... So I'm on a paid subscription, but you can name the project and be like, I don't know, CV or job application, right? This would be like our project. Then, Let's do that here actually. Subjoe applications. And here you create the project, the goals, and something that this will be key for agents to help you get further is the memory and the files that build the context. Because if you don't have those memory or those files here, it will elucidate a lot of things. I noticed trying to get this up, that if I didn't say use my cover letter or my current CV, it elucidated my name, like my last name.
Yeah, yeah, definitely, yeah. This is there for you to use, yes. Yeah, that's a good question. What else, so... This would be like a way to do it. So in a chat you can explore like options even being and I will talk about this a little bit later, but so most of these agents work better when you use really deep contexts about any job that you wanted to do. So for example, you may want to have Like, what are they?
And here are some of the skills that it's... So the question was, what are the skills needed for an HR agent to review, comment, or improve my CV, right? So the first question shouldn't be, and this is important, the first question shouldn't be, this is my CV, improve it, right? Because we tend to do that mechanically because it's easier, right? To not think that much about it and just go to have like, this is your description maybe, and this is my CV, tell me what is missing or I don't know, or improve it to fit.
or give me a cover letter, right? I was doing that for a long time and then I realized that First, agents or these LLMs are programmed to agree with you, right? If you tell them something, they're going to believe that this is true and they're going to actually implement whatever you think. So the value of this space is to have different perspective for how to ask those questions, for example. But one thing is, okay, what do I actually need?
And experimenting with these things, you will see that if you don't know the specific word or the specific architecture or you don't have the knowledge, it won't give you that if you don't ask for it. For example, you have different, completely different results if you say, do this in a GitHub and then let's create this. If you don't mention the word GitHub, he won't say that we're going to use GitHub for doing that.
And that could be a good practice, right? So I think if we all, came up with these-knowledge gaps that we want to address, this is the best thing because the prompt could have like those specific things that you need. So this first prompt was, okay, what are the things that I really need for doing those, for the skills, right? Yeah.
That is a good question. I think, so I'm, as I say, I'm taking a course that like gave us access to the pro version without limits, but you may want to to save credits on using like this smaller models for like not complicated tasks. So you could use like for example, for prompts or for questions that are not complicated, you can use the free versions. But when you use extended thinking, this will take more time. Like it improves the answer, but it also will take more time and spend more credits.
edits in the free version or in the pro version if you don't use the correct model. So my advice on this will be-my personal advice will be on the first plan that you do to use the Opus form. 4.6 that is the most capable of the models. And then when you say, okay, let's implement this plan, then use it like a lighter version of that. And that could help you save some credits there. So, okay, yeah.
So, How I'm using this? You can, okay, we will build this thing. So For example, this is a prompt template. So I'm applying for a specific job and need help telling my resume and writing cover letter. I will provide my base resume and job description. Please, one, identify the top five keyword skills from the job description. Two, suggest specific edits to my resume bullet points to match those keywords. Three, flag any gaps between my experience and three requirements. And four, write a tailored cover letter, three paragraphs max.
Important guidelines. So if you want to set up the context of your responses, your prompts, right? You can also provide materials. So for example, my base resume is this, the job description, you can find it here, right? This is the additional context. Why I'm interested in this role? Give a reason. Why are you applying to that? So this actually, this second layer of consciousness before applying to something, I think is useful for even for you to understand, hey, why am I applying to this?
Am I a good match or not for, and that will help you, like these agents or these interaction will help you understand that better. I'm actually a good fit for this position. If this probably is not the case, you won't spend time in the second process that will be like teller your resume and your CV, right? Your cover letter, sorry. So then you'll iterate with that, right? You can make that a skill, for example, to convert this into a reusable skill, you just create in your folder, you create that cloud skills resume tailor and the skill that MD.
So everything in cloud and I think this is the trend for other chat GPT, other or other apps that are out there is to have these MD files. This is markdown files that you can open on Visual Studio Code or Courser or whatever. And these could be like the skills that you can have. Um What I was So there are some other, Um Okay, let me show you because There are some paid or even other projects out there, for example, this AI Apply.
And what I did there-and I think I can improve that even more with what we've been learning in the last couple of weeks. For each job description, you want to create like a folder. Right, what if like a folder for, okay, so this is my My team. So you have here resume applications. These are the applications that I find out there. So that process, after you find like the one application, what I do here, for example, is to create a new folder, right?
So a new folder, for example, I don't know, this. Um... this one, this job application for this thing. Then I create a markdown with the description. I think that require requirements. And then I just copy paste. This is like Visual Studio Code, like it is the same. Maybe for you to not get used to just one and experiment with other interfaces that like this is as I say anti-gravity but there's also a cool sort of this will still code. It works the same.
So I just copy paste in the requirements.md, I paste the job posting, right? So this is what they're looking for. Once I do that, there are two things that I basically do. So if I'm using the IDE agents, so they have here the agent or the copilot or whatever you have on your IDE, you can use that to help you create the core layer. But what I've been using at this point is the cloud terminal. Um... So here it basically in the terminal it opens clot.
I say, okay, clot and here it has access to these agents. So there's, it is like a .cloud folder inside the project that is my CV application setup. And in those agent folder, you have three, you can find three. Like I was experimenting with how to create these agents, but I have three agents so far. One agent is this career advocate. The career advocate, and I think this doesn't look well on the screen.
So maybe let me change to-a light team that is better for joining. Yeah. That's a good question. So you can create this on your own because it's MD files, but cloud is really good creating their own agents. Let's create one, for example. I'm going to show you what I have and maybe what I'm missing. So I have here a career advocate that is like, you're a career advocate, an elite career strategist specializing in supporting mission-driven professionals transitioning into technical roles.
Because this was my, like, it has to be personal, right? transition from more from an economist role to more like a data scientist so you're working with like okay with me this is knowledge about me i'm an economist with the mastering blah blah blah this is the background career goals so you want for your first setup you really want to go deep and go deep saying ask me questions what do you want to know about me what do you need about one of me what do you need about uh how to i don't know how to be a good match what i've been looking for in a job posting, right?
And that could be at the end like what drives your process to create like these specific agents that help you do things. So your responsibilities as an agent is CV tailoring, optimization, job search strategy, application material coordination, interview preparation, professional positioning. So this will make your life easier, right? If you give a good promo and it's supposed to do some examples could be your career coach and harris for example or here on in the computer science department also it could be something like that okay so this person is and you can be a specific is this this my coach for example in real life is someone that has 10 years of experience in hr and for technical roles and when you do that cloud fills all of these gaps with the knowledge of you or yourself or the project so let's let's try to createWe have a question.
Yeah. That is actually the most important question. Yeah, that's important. I think there's no Golden rule here. This what I'm showing you is not going to be like a foolproof workflow for you to get a job, right? I think it's a lot of experimentation. And what do you think is best? So What you can read up there is that if you can be as specific as you want for your agents to have like these personalities, and I know Chema is there, I want to pass the ball to Chema after this.
So as deep as you can go with these specifications, the better. How do I know what I need? That's important. For example, for a CV or a cover letter, we know we have hearing harries or like we have person that look at your CV and give you comments, right? So if you want to improve that process but not waiting for someone to be available to do that, set expectations and you can be as broad as you can for when you're creating an agent, but also you can ask him what else do I need for this to be a good agent or be a good skill that my agent can use.
So here you can create, for example, an agent So many managed agent configurations. And this tells you, these are your current agents, right? Career advocate, data science career advisor, tech career coach. For example, you tell me what agent do you think might be I'm missing? Because I have like this specialist that is... The career coach is the one. This is the most capable model. And this is like my tech career advisor, especially in data science, because I wanted to apply to those jobs.
And maybe this is a separate thing, but like in your job process you wanted alsoSome agents help tailor your portfolio. Not that you would change your job, but just like--That's good.
Giving you advice on--On my portfolio. I think that's a really good use case of this. Because yeah, okay, let's create this agent to be as specialized as possible in portfolio reviewing, right? Okay, so you... are an expert On. reviewing portfolio and improve portfolio page and improve Add into something Done. 10 sell well not sell videos that like I can show my skills for I don't know, job posting.
Yeah, you can tell for example the cloud to create a markdown file for an agent.
For example, for the free version it could be something like Do a prompt in the chat that is like you asking that to create your first MD agent. And ask question like what skills should this agent have? Like even if you don't know, that will help you like brainstorming on it, right? So okay, so you can show me skills for the job posting that I need to choose. If it's portfolio that is out there, like it's in the web, you are an ex, you have deep knowledge, for example. Um...
Deep, known, ledge. around HTML So you need these words of if this is a portfolio that is out there, so it's a web page, HTML is important, web development. But also design, right? Development. and web design What else do you think that is a good agent for portfolio reviewing? What could be it? web design I don't know, HR specialists? You can be specific in this company. I don't know. I want to work in Uber.
You can have a set of experiences and the model will interpret that as trying to fill those gaps with some backgrounds. from I don't know, civic tech companies or whatever, form tech companies. Um... you Give. Advice and For example, you can have the action. What actions this agent will have? Just give me advice. I don't want it to be touching my portfolio with code, so I just want advice. And I don't know.
So one thing that I can share with you if you are interested in this is that I changed this template that Harris uses into TeX that is the LaTeX file and the LaTeX file here is designed and I spent like two hours with all the agents like to be this should be exactly So that's a replica of the Harris site because it was supposed to be optimized for job searching, right? So once I gave that one of the agents here the task to create that, it uses my CV that was, I'm gonna show you how it was before.
So it was this one. This was the one that I used from Word. And then I created these versions of the job search. from all pure LaTeX and it looks the same. It looks exactly the same as I wanted. So I can share with you this tech and you can say, for example, one of your agents, so okay, I have this tech, I'll add my CV in PDF to this specific tech. So you can have your own CV on LaTeX. IYeah, it is like once it decided for one job application to change your CV, it won't be creative on the format, on the space or everything because it's already, LaTeX is basically already telling you how it's going to be the spaces, the font size, whatever. So this is already fixed as code.
And yeah, you won't need to specify much things on the, but on this side so What I do is, I was showing you, okay, I create here the requirements and once I create the requirements, I say like, okay, now, create a cover letter, Based on my CV that works for this specific for this specific job posting. Then it will create this tag also. and This is a tech that I also spend some time tailoring to my preferences, but it's basically one page.
It's all my knowledge because the cloud has access to all my folders with all my other CVs, with all my other cover letters. And here you're gonna show And you can go personal, like you can say, okay, no, this is too generic. I want to make this cover letter more personal. Use these projects that I have in this folder. So if you treat your folder or your project as a folder with more folders, it can access other parts of your cells.
For example, okay, my specific projects are here in this project.md. And this is like a md.md. Cloud also helped me create that is a highly detailed description of all the projects that I have participated before. So it can select or chooses from all of these different things to make the cover letter more personal or more detailed and do not loosen it that much. And Okay, I think Yeah Yeah. I actually did that and I want to share how I did that but I want to also, maybe Chema is out there.
See ya ma! Yes, I'm here. at the beginning, we were using this brain. All of this information is in Markdown files. So, As more detail that you can have about yourself and create like a huge data set about you, those cover letters, those CVs are gonna be better. So for you to imagine this, it's just one minute that I take to just have one description. Then I say, okay, based on my cover letter preference, create a new cover letter for this job posting and tell me how I need to change my CV a little bit more for showing the skills that they need.
It will do it on their own. Now you have these two PDFs to upload to that process. So I'm spending more like five, 10 minutes into one job posting. So this solves one bottleneck that is, I'm going to apply to this when I have time. No, you can just leave that working and then do other things. And then when the PDFs are ready, you can just send them out. Yeah, Tema, go ahead.
I do something really similar but probably a little bit more user-friendly than building your own agents. So within Cloud, you have this tab that is called Projects. And within projects, you can just create your little, basically your agent with some basic instructions. and also add some files, right? So usually what I do, like I have this, I have this project created with this instruction and it's basically, you know, like the, The job is to land a job, act as a job hiring manager of that particular job.
So, Cloud knows that it's, you know, I don't have to personalize it for every single company. I give them what company it is and it will take the data. And then... The idea is that it evaluated from 0 to 100 on the likelihood or landing this DATS resume. A review? And then I give it some templates. So my idea here is I want Cloud to use at least a ton of tokens. So if I create some templates and I say to it, hey, use this template for this project, it wouldn't need to search through all my information in order to every single template.
It would just use one baseline and then it would require less tokens for every single computation that it's doing after that. That's just a human decision. I think this is the best one. I had different, you know, data engineer, BI, public policy. So, yep. And then the way projects work is that every single new chat will just use these instructions and behave like that for every single chat. So every chat is a new job application.
For example, you can go, you know, this one, Morningstar, okay. There's one... So... As you can see, it's taking the instruction I gave. I just give Cloud the job application text, And it ranked my likelihood. What am I missing? What am I doing right? What am I doing wrong? Some things, you know, some suggestions also. Other thing is I explicitly told Cloud that, "Hey, don't fabricate an experience.
Just give actual recommendations on what I can do based on my experience." Because it's really prompt, like I have realized, to just make things out. Like, oh yeah, you used this. And I was like, no, I didn't. So be careful with that. And yeah, this is the way I usually use cloud, you know, with this project. So every single chat is just a new, just using these instructions. But as you can see, you know, he gave me gave it to me in a little clear, concise way where I can like scan what am I missing?
There are some times where the It just looks like text. So it's not... Yeah, it does not have a... like a template to use to tell me what I'm asking it to. So what I'm actually doing right now is that I want to create a template. I'm building like this website just For my requirements, I want to keep a job tracker because my job tracker is just in a Google Sheet. So I want everything to be in a single place where I don't need to change tabs.
And I created this agent where I can just tell Cloud, like, hey, use this resume. And you know what, like this is the, Now I don't even need to copy and paste the description. If I want to, I can just send it a link. And then Evaluate will now give me a template that I know what I... Based on time, I realize what I want to know, what I don't really need to know. And this is a way to just standardize that and keep my workflow going.
Then after that, the idea is that This will just create a database on its own on these job tabs that I haven't created. So I don't need to also keep a job tracker, you know, on the side. Like everything can be done in this same place. And that's, I don't know if anyone, I can't hear the groups. I don't know if anyone that was interested.
I think this is really cool when Chema showed me this for the first time because I think that score is pretty useful and that's the first time I've seen that. It's like, how is it Evaluating that, what is this ATS thing or what was the problem that you told them to use this scoring technique and the gaps?
Yeah, sure. ATS, for people who don't know, is like the automated system. FHR uses this to filter resumes. So I use this, you know, most of the jobs I'm applying are just cold application with a recommendation. So the biggest filter actually is going to be like the first, the system, you know, after that, I think I got a shot. So just to maximize my likelihood that I go through that first filter, I told Cloud that I use, like, VDH is like an ATS old.
I think that's pretty cool because That's probably like, and the same, like I was saying at the beginning, like, so these models are designed to agree with you, but if you design a mechanism that evaluates you and gives you that score, like as it is, like then you can actually improve, right? Like, because if not, every, if you just say like improve this CV, he will tell you what you want to hear at some, like at somehow, but now this is like a better tool because it's evaluating your gap.
more honest on the evaluation system and then you can actually see what's out there and not just the problem of improve my CV or improve my cover letter but this is just telling you about your own skills and I think that was pretty cool, Gemma. I don't know if...
I've been using these for the last two to three weeks, so I... I think maybe too soon, 'cause just to say like, if the rate of approval has improved, However, what I will say is this definitely made me more productive on applying. You know, if I was applying to, I don't know, one job per hour, maybe I can do like two odd jobs per hour. Yeah, I don't spend as much time as editing or thinking about what I should have in the resume.
And then you can decide, okay, this is a problem of wording in my CV, or is this a problem of I don't still have those experiences, right? And if you create a system that fits or you have the data about that, you can improve it. And it will improve on your actual metrics and your actual outcomes. And I think that's the best way that you can use these things to actually help you recollect that information that is costly to fill yourself and then make it easier, right?
Making it explicit like theseSo the job description resume pairings that I'm getting interviews for, Yeah, I'm still missing for example that part because I just I'm playing but then when I don't get an interview or whatever I'm not telling them all that I'm not getting it. So I think I may want to start doing that and then these models will learn from their errors or from their like to their trend to agree with you and then kind of change what you need to be changing.
Thank you, thank you Gemma, thank you so much. I think that was wonderful and I really like your setup. Thank you. Thank you. Okay, so, I think for the rest of the minutes we have left, like 10 minutes because this was heavy, maybe two things. So maybe, yeah, maybe the next session will be even more basic going earlier like before. And if you want to leave that, I think that will be, I think for us to get more people, we need to be more Probably more honest on the schedule, right? On, okay, please fill out this availability if you really want to go to the session. And I mean, even at one point, no one is feeling that, that's okay. But then we can also promote that earlier in the week because I think that, yeah, all the TAs on the Friday has been complicated.
But I think that session could be, and I don't know what other people think, but about, okay, like really from zero and how can you do that? And maybe people can come out of this session with a personal setup that works for them. I think what Chema was showing is super cool because it didn't require to have the Visual Studio Code and the other parts. but still show how to use it with what is available there and this is even in the, it's the pro version of Cloth but still not the Cloth Max. What he was showing at the end on the Job Tracker app, he's building it with the Pro Max Cloth with all the tokens.
And most people we talked to in the first session were not paying for it. Well, actually, there was a few. Maybe we can, in that poll, we just collect what subscriptions are like and if most people are on free accounts we focus on tutorials at least the next one that you can do like fully free that's great Popping glow?
Yeah, we still want to do a session about OpenClaw. And I think James was... was I think excited about doing that but in a way that is open like all open source right is like all should be open source and so I think to get there we probably need more experimentation with what we have now to understand what we want from the free models to get to.
A good amount ofLike, I think risk associated with connecting all of yourYeah.
So maybe that's like the session into which could be about like security, privacy and how to, Yeah, how to do it? And he was super open to let us use his servers. So he's paying for servers in, I don't know, Norway? It was Norway? I don't remember.
Asking them to create them. So I did this also as Shemama was saying, as an app, a job search tracking to 2026. So it's supposed to have a database, it's supposed to have Um... scrapers from greenhouse, from base, from other static scrapers. It's supposed to have like some models. So this is the readme of the other. That's all? That was all?
I'm trying to see how far can we go with this. So this is our automated web scrapping and tracking system for job opportunities, special data science, machine learning, my personal interest. But I think this could be also in a workflow that helps for everyone. I can share also with you how to do that. So things that I use to create this, I was telling this app to use React. Why? Because I think there's a lot of things, I've heard this word, I can actually don't understand how React works, but-That people use it, right? It's like the app, the web app front end and how it's connected to the back end and everything, right? So we need memory, we need bad databases. And Docker, because I learned this from Data Science Institute, they use Docker to contain all the applications to test it and then you can actually deploy and escalate with that.
So it basically scraps the sites that I was telling them to go to. So get the information, get the dashboard of job offers that are there. And if I deploy this, it should create alerts. So, okay, this job matches your CV with this percent of things. And then it can track that I... So the part that I was saying, I'm not getting feedback on if I got a response or not. So it's a job tracker that actually helps me get the data I applied to this, I got this, an interview or not, or they, like, I didn't get to the next phase. So it's trying to use that.