Vibe Coding? No, Vibe Marketing!

Who here is thinking at all about getting content, visible on LMs? Yeah? Kinda SEO, g e o? Do do we have a yeah. Okay. So I was on a, I was on a I was watching a YouTube video of some experts. All the experts in the industry were talking about how you optimize content for LLMs. And the words cosine similarity were said multiple times, and I was like, I don't I think I know what that is kind of ish. And so I took the transcripts of that video, and I kinda reverse engineered what they were saying and took some screenshots. And I made myself a little, product requirements document, and I started to vibe code what I'm gonna show you tonight. So, it's called Symcheck. What's on I guess, similarity check. I don't really know. What it does, it takes a topic, and it takes a URL or you can paste content in, and it will tell you how similar that content that that topic is to the content that you've written or that you've that you're analyzing. And that, in theory, will help you get it visible in LMs. You have to really get cited and other stuff like that. But, basically, it's it's it calculates a cosine similarity score. It's not academic. This is real practical. Like, I'm actually using this with a client right now to see if I can sort of make a difference in their content. How I built it? Mostly Claude code, although codex is awesome. I've been using that a lot. I was joking on the way over here. I was talking to somebody. I I, wanted to check the code today with with, 4.6, and I think I might have sucked down all my tokens in, like, 1 like, can you just check this? It's like, oh, you're out of tokens. Great. Awesome. So, what you see on screen here, and I wanna just show I have this project template. I think I've I don't know. I you do watch these things and you, like, watch videos, you're like, this guy described this process that basically is almost like a reverse engineering, interview process. So you, you set up the project, you drop this project template in there, and you open this prompt, the planning prompt, and it basically, and I'm by the way, happy to share this to anybody. I'm on the site. So if anybody wants this, I'm happy to give it to you. It goes through, like, question by question, and it asks you all the questions. Now, again, I'm not technical. So a lot of times, I'm like, you tell me what do you recommend. So what this thing recommended was a couple of things. 1 was 1 of the design constraints was, I wanted this to run locally because I'm tired of paying other people for things, and I've been playing around with, I have LM Studio running on here with Dolphin Mistral. Now this is a MacBook Air, so I can't run at 16. I can run it, but I can't run anything very big, but it's good for testing. And then I, what I've been doing lately is using OpenRouter to sort of flip the switch so I can go talk to full size models anytime I want. Right? So local first, it wrote a Python, application. The first version used what turned out to be kind of a very weak sentence (Terraform), if anybody knows what that is. So I upgraded that. And so at the end of the day, what I have and what I will show you now is, is this this thing I call, SIM check. So now it's running locally, although I did put it on, I put it on, streamlet just because I was like, oh, there's 1 button. And I like, again, I'm not technical. So I'm like, oh, I can host this? And then so this is awesome. So, again, if anybody wants a URL, happy to give it to you. And it's evolved. This is, like, version 3 or 4. I've upgraded the model, and I've started to show it to other marketers and strategy people who are now giving me feedback, like, what does any of this stuff mean? How does it actually work? So there's actually a start here kinda screen that tells you what you do. It's just super basic instructions. I'm gonna actually walk you through it so you can see what it does. So let me just grab URL here. So I found a, a URL that is, a pretty good article, ironically about, answer IDENTIFICATION. I just see now it failed. See, that's that's I'm gonna I'm gonna let's just see if this works. Otherwise, I'll just give up. Because the other mode mode is I don't need to use the, I don't need to use the, the fetcher. I can just paste content in here, which is good too. What is your what is your name, sir? Sam Fleming. Let's do Sam Fleming. Okay. So the whole idea is you kick a topic, you take a bunch of content, mash it together, and see what see what the cosine similarity is, and then it will give you a set of action items. This thing has a bunch of different modes. So if you wanna it it'll auto detect the content, but also if you're doing you wanna paste markdown in or HTML, you can do that. That's all fine. So if I run analyze document, please work. Please work. Oh, it's working. That's working. Okay. So it it will tell you a bunch of stuff. Right? So there's 7 chunks on that page, and there's some defaults that that you can set. And this score is an invented 0 to a 100 score. So this apparently, your about page is 11% is 11 out of a 100 for cosine similarity. What the application does is it gives you a literal action plan. So it will say, you know, add a crisp definition near the top. So, like, let's just go back to that page. Does it say your it does say your name at the top, so it's a work in progress as we've seen so far. So it gives you action. It gives you action items. If you feed it code, it will look for schema, which which will help. And then the other thing that's interesting is that it gives you, all sorts of tips on where to look to improve the content. Right? So there should be a loop. Now what's cool about this is if you wanna just continue to edit the content in here, you can continue to Apple’s, and it will just give you feedback. And when you're happy with the content, put it on the website, which would be good. Within here, there's also a bunch of different settings that I set up, which is, let's say you're running an ecommerce site and you analyze a product page. Great. Or a how to page or an FAQ page or informational page. So basically all the different modes in theory that you could use to improve. I'm gonna guess the fetcher is down because I I used, I found this site called URL to markdown, and I talked told my friend Claude, hey. Can you just, like, make an API and just so that I just hit the button? So that that must be what's down because the rest of this is working. Okay. And then the last piece I will show you and I will disappear is there's a statistics section in here, which again is kind of interesting. The other thing I'll say is, like, part of why I built this was for me to understand and learn this term, you know, cosine similarity similarity. And so it built me a whole bunch of ways to think about this stuff. And this is my favorite part, which is it looks at all the trunk all of all the chunks from the content, and it tells you, basically how how they rank. Strong, weak. And so you can start to really go through the content 1 chunk at a time and start, okay. I need to approve this here, this year, this year. I'm not sure. I probably had a time. That was insanely ridiculous. I'm done. Thank you.