Graph Engineering in Practice: An Agentic Workflow That Tailors Resumes and Applies to Jobs

Speaker 0: And then, so today, I'm going to demo the using of the agents to build the each page intake for the workflows. So I've been doing doing this for quite some time. So 1 of the workflow that I would do the demo is how we can uses the agents to build, workflow to be around, like, job application. It sounds really very simple in the workflow, but when you try to do it, it's so tedious. There are, like, so many new AI. Speaker 0: You go through all the company websites, figure out what's the right job description based on your work, you tune your resume, and then eventually, you know, they apply. So I created another workflow just to do this. For me, the idea itself is that because the agents are capable of doing a lot of, AI, there's a lot of existed. From my point of view, it's, really very simple. Essentially, it's nothing but it's a Code edges and between of the 2 nodes, we just share the information. Speaker 0: So with that, what I do only know we have is something like this. It says open source on the project. You can check it out. Speaker 1: If you Speaker 0: see okay. So I created a graph, which is nothing but it's just the pipeline, what you would normally do if you are using agent. So you're manually looking for a job. So, essentially, you go to the website, you draw the website, you score the job on the description based on your work, you create your own, I would say, let's say, the resume. That's why we do the data and the pretty. Speaker 0: And finally, you know the entire. So in the back end, I use on the Google, so the ADK. There's a lot of, the other open source frameworks you can use. There's strands. Even Postgres has its own AI need to do is create a watch list, essentially, the list of the query on the website. Speaker 0: You can just create the watch list. Essentially, the list of the query on the website, you can just ask your any local agent, codecs, cloud, to create this conflict file. And the only thing you need is your c, which I call it your, let's say, your base and your resume. Over here, I use some pipeline. That's it. Speaker 0: At that point, I have a small demo. So this is how the discovery really works. The system would not load up. It it starts in this, crawling. The website's over here, figures out what's the right control for you. Speaker 0: Singh, yeah, heaps on the right. So, yeah, so it's a very simple workflow. The idea itself is that, agents have good enough new models improving them day by day. So if you can really create a good workflow, Speaker 1: Yeah. Right. Yeah. Very much. Thanks. Speaker 1: So, Max, if you could step up, I have a couple of questions for you, Sayantan. So so to to be clear, you've built, an open source agent to apply for jobs. Is that right? Speaker 0: I would say it's a sequence of agents. It's not just 1 Sequence of agents. Speaker 1: Yeah. Okay. And it's an open source repo anyone can check out. Anyone can check out. Out. Speaker 1: You can run it with your local codecs or cloud or you can use any API key. Yeah. Yeah. So I AI tons of people approve me Ask how they can find jobs in this market. This is a really cool way. Speaker 1: AI we work also talking about the Workflow, and you were saying and and if you put any of that graph workflow in code base, so if Speaker 0: you Yeah. It's in the code base. In the code base. Code base, like, how it uses some sequence of agent, for example, when it's building your resume, does review and critique kind of workflow. Yeah. Speaker 0: Build something, the other agent then validates that. It matches with the facts that I had not initially placed. That's why it's not leaking up information. Got it. Speaker 1: So it's Speaker 0: I've got a question here. Yeah. Speaker 1: So I I also made something AI similar a while ago, to look for jobs. But, 1 of the things I noticed was, I was approaching it through a kind of a different angle AI I was basically going, lead, I wanna be notified for new particular company as soon as they come in. Do you notice, for example, you know, maybe kind of allowing the Agentic workflow to build out, you know, for example, scripts that would say, hey. This is AI a hard Code rule set that will give me, like, a good enough heuristic to identify if a job is related or not AWS opposed to letting the LLM run all the time and, like, run your presence. Do you have you noticed, like, that kind of a work? Speaker 1: So I Speaker 0: do have that kind of work. So the idea itself is that I run it, like, every day at midnight. So So it crawls for check on the website. And then based on the description and based on some rules, reference that I've set, it tries to score and figure it out. Okay. Speaker 0: This is something which is good enough to proceed, then it goes uniform. Speaker 1: So is this kind of file a timed run that apps? Speaker 0: It's right now, it's another AI. Again, it's I SDE it as, like, a very simple only corona of the timed run, but, again, it could be based on any events. Cool. Yeah. Speaker 1: Thanks. Alright. We're out of time. We've got 1 more