I Started A Coding Habit & You Should Too...

Key Takeaways

Today I want to do something slightly different. It's Wednesday. I usually don't publish on Wednesdays, but I want to do some extra videos, so to say, where I just talk about stuff. I ramble. I just yap about something. And specifically, I want to talk about a habit that I recently started. It's a daily or almost daily coding/studying habit. So, I basically work on stuff that doesn't have a specific goal. Now, I already code on a daily basis without this habit. I work on a startup. I work for freelancing clients. I also prepare code for my videos. But this habit in particular is not goal oriented. So I'm not building stuff to achieve something. I'm not trying to build a company. I'm not trying to make money. I'm not trying to prepare videos. It doesn't have deadlines. It doesn't have requirements. I'm just coding for fun because coding is something that I like a lot. And when I do it with a purpose in mind, it is still enjoyable, but it's more about achieving something than coding for fun. So yeah, this video today will mainly be about this thing, about this habit. And maybe some of you guys find inspiration in this. Maybe you want to start something like this yourself. So let us get right into it. >> All right. So let's talk about this. I want to maybe tackle this from a couple of different angles. On the one hand, I want to talk about what the benefits of such a coding habit are. On the other hand, I want to maybe briefly describe what I'm doing and how I'm approaching this. And then maybe I want to give you some inspiration for why you should start something like this uh yourself. So the first thing is what is this habit that I'm doing? It's not really a commitment that I have to code every single day. It shouldn't be an obligation. It shouldn't feel forced. It's more like a try to do this almost every day to code for like 30 minutes, 60 minutes. And by coding, I mean do something that you're interested in for fun that doesn't have an immediate benefit. Now, of course, every side project that you do will have some benefit down the line. You will become a better programmer. You will become a better engineer, uh a better business person, depending on what you're doing, of course, uh you will become better at what you're doing just by playing around with stuff. But you're not doing it because you need to do something by the end of next week. So what I'm currently doing in particular is on the learning front, on the studying front, I picked up a book that I already worked on a couple of um yeah actually months or years ago, which is this one here. This is not an advertisement now, but it's this hands-on machine learning with scikitlearn and tensorflow. Now, I'm not doing this to learn TensorFlow. I basically skip all the framework specific stuff because I think nowadays you should be uh learning PyTorch, not TensorFlow. But I still want to finish this book because I started it at some point and I think it's interesting to recap the stuff that you already know. So most of the stuff in this book I have already played around with. I know the architectures. I know how they work. I know what to do, how to pre-process the data. But it's nice to, you know, tackle this again from a more theoretical perspective. I also have a book on mathematics for machine learning. I don't have it here now. It's uh in my shelf. And I'm basically going back and forth between these two books just learning stuff because if I now learn linear algebra, it's not that I really need it for my next project. It's not that I really need it for my next video. But having good linear algebra skills, it's just nice when you read papers. And of course, reading papers doesn't necessarily solve a problem right now. But if you read papers on a regular basis, if you study math on a regular basis, you're just more competent at what you're doing, especially in an age where everyone is just vibe coding. Which brings me to the next part uh the side projects that I'm currently working on. Now first of all the inspiration um not the only inspiration of course but I think all of us know uh Peter Steinburgger he's basically uh the guy behind openclaw he's also Austrian and um this guy inspired me not because of the success that openclaw had but because of openclaw I found him. I went to his GitHub and I saw all the projects that he had here and I saw his activity graph and I felt inspired by that because I thought look at uh the years that he has been coding all the time now mostly on the same stuff nowadays mostly open claw but it's just interesting to see how many projects this guy has implemented to essentially scratch his own itch to do something that he would benefit from he built openclaw for himself for fun and I wanted to follow this example and just start building stuff without having a goal in mind like making a video about it or selling it to someone or making a product out of it. Now, not saying that it can't happen doesn't mean that one of these projects cannot become a company but I just wanted to start doing stuff for fun. So, I basically started these two projects here. One of them is called latent or latent assets. Uh very very basic example. I think I can run this locally here. So, I have a directory on my computer now. Site projects in here. I have uh I mean I started one dragon torch which I dismissed because I didn't like the idea too much but uh latent assets basically what we can do here is I can go into this directory experiments so the experiments then I can say in the inference API that is just a multimodal embedding model so it combines the embedding space of images and text. So I can just feed in images and I can also feed in text and they're embedded in the same space. Um, so if I say uv run fast API defaf and then port 99999 and then I open up another terminal, I go to the same directory here, but I go into the project directory here. I have backend and front end. So I can go to back end backend uv fast API def and then a third terminal final terminal which is going to open up the um actually what happened here? Yeah, of course I need to provide the the file as well which is the fast API server. Forgot about this. Uh in this case it worked because the file is called main py I think. Um and then if I go here to the front end directory in latent assets I can go front end npm rundev. It's all in a very experimental version here. But the reason I built this is not to show that I have some skills or to to build this for a client or something. I'm doing this because I genuinely need this. If I do thumbnails, I I usually do thumbnails in Canva. And Canva has this sidebar where where you can just upload your assets like logos, the Python logo, the neural 9 logo, images of myself. I can just upload it there. But every time I upload it, it's a duplicate. So I can upload the Python logo for the 15th time and it's going to be its own asset, its own resource. And the moment I try to do something with for example, I don't have an easy database. So if I want to use a Python logo, what I have to do is I have to go to Google, type Python logo transparent, get it, copy it, and paste it into the tool. Now, I'm not sure if I have images loaded here, but if I type Python, uh I can get to Python logo. Why can I get to Python logo? Because I have a database locally that I've used before where I have a bunch of uh assets. So I think I can maybe increase the number to be uh larger. And you can see all the stuff that I have here. So most of this is thumbnail stuff or uh my wallpaper that I used before and some assets on my on my blog that I'm using. And these are not manually annotated. So I'm not actually going uploading them calling this Python calling this neural 9. These are just images. But the embedding model since it is intelligent, it's trained. Um it's running locally by the way. And if you're interested in that, I can also show you that this is the I forgot the name. I think it's called something with Juna something like this. Let me see what the code says. Uh it's Gina clip v2. So this is a multimodal embedding model. I basically have embed text embed image here. And the idea is that the model itself can realize that if I type something like neural 9, even though it doesn't know the brand neural 9 most likely, it knows that this is like a brain and a nine. So it figures out that these should be listed in the very beginning. And if I type something like car for example, it's going to take this here at the very um in the very start. And if I type something like microphone, it's going to probably get me this here. And of course, I can also limit this to be only five images. By doing this, and I also have a feature if I click on this, it's in my clipboard. Um, it doesn't show a message here, but if I click on this, it now has been copied to my clipboard and I can just easily paste it into and work on the thumbnail. And I can also manually add text. So if I say let's say my company is not called neural 9, it's called uh I don't know FD intelligence because Florendov is my name. Intelligence would be the thing. If I now type FD intelligence because that is my company. I mean in this case intelligence is linked to the brain. Let's go the extreme example and say my company name is snake car and the logo is the neural 9 logo. If the embedding model cannot know that it cannot figure that out. What I can do here in my tool is I can go and annotate this with snake car for example and I can just say safe and I can maybe do the same thing here. I can go and say snake car. And now this is a hard-coded tag. So if I search for snake car these two are going to appear at the very very beginning. And you can see it is because they're tacked, not because the embedding was matched. Um, so this is just a very useful thing. Also, if I go and upload something like let's say I have here some thumbnail, I can just upload it. It's now part of the image uh thing. And I can also give it tags if I want to. If I don't give it tags, it also works. I can just go and type PyTorch now and it will hopefully find this thumbnail here uh very early. I'm not sure why these are considered more relevant here in this case, but yeah, it's not perfect. You can also assign tags and you can fine-tune it. But the thing is I just have a tool that helps me and this implementing this was a lot of fun. Now speaking about vibe coding earlier, how do I approach this? Because of course I'm not typing everything here by hand. So in this latent assets thing, what I did specifically is I played around with this by hand. Playing around with the inference API, playing around with the model here basically. Okay, how do I send stuff to uh to the inference API? How do I build uh how do I interact with this endpoint? Then how do I work with the vector store? Then I downloaded the cipher 100 data set and played around with this. I did all of this um manually for the most part. And then once I figured out the logic, once I understood what I'm doing, once I understood the use case and how this actually works, what I did is I fed it into claude code or into a similar tool. I'm not sure what I used here. I think it was claude code. And I told it look I know how to do this. Here are three files that I wrote and I in these work basically do the same thing as it's done in these three files and build a fast API application around it for the inference so that I can later on connect it to different APIs and also use Qrent uh load it into or save it in a file load it from a file and all that and then build a UI around it and then I specified what kind of UI I want to have. This is how I built this. And this was uh work of I don't know, it wasn't a lot of commits. It wasn't a lot of work. I did it in two or three days, I think. And I'm still continuously pushing updates on this. So, why am I showing you this? Not to tell you what kind of awesome tool this is. This is actually made for an audience of one. This is made for me. Not saying it cannot benefit other people, but this is just for me because it helps me when I design thumbnails. I want to have my resources easily available. I just want to go into this here. I'm looking for the Python logo. There you go. I'm looking for the neural 9 logo. There you go. I'm looking for sec scikitlearn. There you go. I can just copy. And if I have all my assets in here, I can easily work with this. Uh second thing I did, my mail inbox was full on my basic uh email. So not my neural 9 mail because there I have my own yeah my own server and everything, but my public email that I'm using for personal stuff. Um it it was basically full. It has a 1 GB limit and I want to free up space. How do I do that if I don't know where the important emails are? I basically built this. This is also the UI is mostly vibe coded here and also this is an outdated screenshot actually. But the idea is I have this. This is now sample data. Here I have my emails loaded and then every email is a square. So basically every email depending on the size has a different square size and I can just browse it. So let's close all of this here and let's go to male deer stat. By the way, the name is inspired by this software called QR stat. I'm not sure if you know it, but stat is basically or kadier stat for KDE is basically this tool where you can do the same thing for your disc. It's I think again inspired by Winder Stat, which is only available for Windows and thus an inferior piece of software because everything that's only available for Windows is bad. Uh but essentially what we have here is UV run main py a flask application. Yeah. So here in this tool I can basically uh specify now how many the top how many emails I want to look at. So the largest 100 200 emails I can group by sender. I can load I don't want to show this here right now because I only have actual email inboxes that I don't want to share here on video. This here in uh GitHub where is it here? This is sample data and I don't have it here on the system right now. But basically, you can interactively click on the individual mails. You can click on the individual squares and you're going to get the text, the content as a preview here, and you can easily see which of these emails. And then you can also group them, as I said, by sender. Which of these senders or emails take up the most space? And maybe you find something that's irrelevant. Of course, features can be added to this. But the essence of this video today, the essence that I'm trying to to say here is you should start doing something for fun. So these to tools are actually solving a problem for me, but are also interesting, especially working with this multimodal embedding model. I just play around with it. And as a byproduct of that, even though that's not necessarily what I'm doing this for, I now have a bunch of video ideas. I can make a video about CUDA. I can make a video about multimodal embedding. Um, and I can also make a video about mail processing or something, whatever. And maybe at some point this actually becomes professional enough to turn this into a SAS application, into a product, into something that you can sell to people. And at the very least, this is going to be a very nice thing on GitHub to have, especially if this becomes more professional. Because what happens is the activity graph is going to be naturally filled up with activity. It's not going to be artificially filled with some basic script that fakes your history. It actually shows that you're consistently working on stuff that's interesting. And then you can show that to clients and you can build a list uh like Peter here with all the different products. And this is already a very impressive thing for employers, for business partners, for freelancing clients. It just shows that you're proactively constantly working on stuff. But again, don't make this too goal oriented. My suggestion is do stuff that you're interested in. If you don't know where to start, just play around with something that you find interesting. It doesn't have to be useful. It doesn't have to be impressive. It can be a hello world in Rust if you want to learn Rust. It can be uh something like my site projects here to solve an issue you have. It can be whatever you want it to be. But my suggestion is to all of you guys if you have some extra time on your hand, reserve 30 minutes every day to learn or do something in the coding space, in the machine learning space, in the engineering space that you are just interested in without having any expectations for outcomes. So that's enough talking for today. Hey, I hope I was able to maybe inspire one or two of you guys to start a habit like this one. I think you can benefit from it a lot. If you like these sorts of more natural casual videos that don't really have an educational value, per se, I'm just talking because I randomly wanted to today. It's a Wednesday. I usually only publish on Mondays and Fridays. If you like this sort of video, let me know in the comment section down below. Also, of course, don't forget to hit the like button, the notification bell to subscribe, all the good stuff here to support the channel. And as always, on my website, you will find also a services tab, a tutoring tab if you're interested in any of that. Besides that, don't forget to subscribe, hit the notification bell, and not miss a single future video for free. Other than that, thank you much for watching. See you in the next video and bye.