Breakeven Brothers

You're listening to the Break Even Brothers podcast, a show by two brothers shooting the breeze about business, finance, technology, and everything in between. All views and opinions express are solely our own and not affiliated with any third party. Hope you enjoy the episode. If you do, leave us a review and drop a like. See you there. All right. It is episode 15 of the Breaking Brothers podcast. Bradley, how are you doing? Doing absolutely fantastic. It's currently Sunday. Been a pretty busy week, but busy in a good way and tons and tons of AI updates. So I've been dying to just have some downtime to play around with things. But, yeah, all things consider it pretty good. How about you? Yeah, good. Super busy for sure. super busy with my regular 9 to 5. So, you know, not like pulling, like, you know, extra long hours. You know, I think I'm kind of past the point in my career where I feel like I want to do that. But just busy, you know, from the moment I log on to the moment I get off. So I definitely, there's been a lot of news and stuff that's happened that we'll talk about here and new things that are getting launched. But I haven't been able to really dig in and kind of play around as much as I would like. But, you know, I'll make up for that here in the coming weeks once things quiet down at work. It's like that time, like new year comes around. People had a lot of goals for last year. You're setting up goals for this year. And I think at work, everything is starting to pick up again where people have some ambitious goals for the quarter for the half. And we're getting to that point where I also feel the same like busy nine to five, like fully booked and focused. And it's good in a way that I like I like what I'm doing. It's very fun. But it's also like less time to like after work, I feel like, oh, you know, less time to kind of play around with things because I have less energy. I know we talked about it in the previous podcast. of getting home at like 5, 5.30, getting to like other activities, like post work. But yeah, I think it's that time of year where people are picking up again. People are back from vacation. Definitely a little bit more busy. So you've got to be mindful of spending your energy. Well, and I think too, especially, you know, like I've been trying to get better sleep. I talked about that in the last podcast. And like I've been like trying to get to bed on time where I think before, you know, kind of, you know, grunt through things and like work on things late at night. And at the end of the day, I felt like after too much of doing that, like you just become ineffective and like the diminishing returns you get on like trying to work hard on something when you're like running out of steam. It just doesn't really work. Especially I think as you get older. That's just, you know, what I've noticed in my own experience. And also too, it's like what I'm really trying to do this year like consciously is like focus on like what's in front of me, you know? And I think it's easy to convince yourself that like you need to be doing. all these different things and like, you know, once you're done with this thing and then you got to go do the next thing. And it's like sometimes you need to just simplify, focus on what's in front of you, worry about the rest of the stuff, you know, later, but don't like try and solve three problems at once. And I think you get better quality of work on like the thing that you're doing in that time. So as I've been trying to do and I think, you know, so being busy at work, it's like I'm just trying to do the thing that's in front of me. And then once like I'm done with work, it's like, okay, I need to like just pause. Sometimes I'd do some stuff on the side. You know, just far as like working on, you know, You know, we talked last time about, like, coding a little, like, chat agent, you know. So those things are fun. But then, too, it's like, I need to, like, work out. I need to make some dinner. And I need to read a book and go to bed. So it's like sometimes simplifying things is nice and, you know, keeps your focus a bit more than trying to do everything at once. Yeah, I think I think Huberman came up with a podcast recently talking about, you know, the usual health guru stuff. But he came out with, I think, a 30-minute focused episode talking about that you don't have to listen to my, like, three, four-hour podcast. But you can listen to a 30-minute version about, like, focus and, like, deep work. And I think he mentioned, I don't know if it's exactly correct, but like a very intense focus session, maybe 90 minutes before noon and like another deep focus session in the afternoon. But after that 90-minute session, your brain is, like, pretty taxed and you don't get as much value out of it. But if you can get your highest priority work done within that 90 minutes, like, that's the biggest bang for your buck. And I've tried to do that. I noticed when I come in the office in the morning. as long as I'm not distracted by any people or like having too many things in front of me, if I just have one thing and I get to work on it and I'm like energized by, you know, waking up and having that large pocket of energy, get a lot of stuff done. I think the hard part is just choosing, one, the right thing to work on and two, not being distracted within that 90 minutes. Because if you're dialed in, like headphones on, no one's interrupting you, like maybe you don't check Slack that often, you can get a ton of stuff done. Like I've been trying to reduce distractions and like reduce meetings, all those things that get in the way for like a deep productivity session. So if your calendar is filled, maybe a little bit harder for you, but a suggestion of just like 90 minute block and just get stuff done. Because other than that, you have gaps here and there and it doesn't feel as productive, which sucks. Yeah. It's an interesting topic that I actually been thinking about, and this is funny that you bringing it up to, is like the concept of like deep work versus like, I think the flip side of that is called, I think they call it like grease the groove. I think that's used like in like a physical sense. I'm not sure if you've heard that phrase before. Basically, grease the groove is like where it's like, instead of going to the gym for an hour and like working out your muscles and like just obliterating your muscles, you know, breaking them down. So that way when they repair, they get stronger. That's what conventional weightlifting is about, right? But instead of doing that, like just every day, you know, a couple times a day, you walk by a pull-up bar and you just do like two or three pull-ups. And like that supposedly like grease the groove keeps, you know, keeps the strength there. It doesn't like tax your body too much, but you still get the strength benefits of it. I'm not, you know, I don't know the science behind it, but like, that's just the philosophy. And I was thinking about that in the context of work because it's like, you know, how long can you maintain like a deep, dedicated focus on something? Because I think I used to be more the opinion of like, give me like four or five hours, like put me in like a dark room and like let me just go. You know, and like it's like the, you know, Dragon Ball Z. It's like the hyperbolic time chamber, you know, it's like, let me go in there and I'll come back out. Get everything done. Yeah. But I think now, I don't know if I'm, I feel the same way or maybe I've changed because sometimes I feel like, you know, give me those 30, 60 minute increments. Then let me go, like, take a break and go walk and get something from the kitchen and go sit outside in my patio for like 10 minutes. And then go back and start working. Like those little intermittent breaks or like take a walk outside, you know, for 10 minutes. I think they kind of keep me a bit more productive, honestly, than like sitting there for four hours and like forcing myself to like get through the things. And I don't think that's distraction. I think that's kind of, you know, I hadn't heard that Huberman topic, but, you know, that seems like kind of what he's saying. It's like, you only have a limited amount of time for your focus, then you need to reset or just take a break and come back to it, you know. Yeah, I think I've tried doing those, like, longer sessions too. And I think the problem that I run into is there are some things like building a split my expenses that there's just like a really hard problem. And it feels meant to like 90 minutes won't give me enough to get it fully done. And it feels like one of those things that I need to just get it done in a chunk of time. maybe it's three hours, maybe it's four hours. And those are kind of the problem issues I have with working like nights and weekends on split my expenses is I have a lot of small problems. And I get to those within like a 90 minute focus period or 60 minutes, even that. But the harder problems, they really make me think. And I don't want to have focused work and break it up in a sense that I lose context. So much happens in life. If I work on something on Monday and I pick it up on Friday, definitely a little bit more fuzzy. Like I'd rather just get it all done, make it perfect. tie it up. And so I think I'm making good progress on SME, specifically with a mobile app recently, but there are some kind of large items that I've yet to crack where I can find the time and how best to make that happen, where I'm not locked in a dungeon for four hours or five hours. Like, I can make it happen on the weekend, but I'm not even sure that's the right way to do it and if that'll be like the best bang for my buck. So I, yeah, I have been making good progress there, but not necessarily on some big ticket items. I just want to finish off. hopefully in the coming weekends I'll try to figure out what that looks like. But yeah, I think deep work is definitely something that matters a lot like having focus time. My manager at PayPal is really into deep work. Like read a bunch of books on it. I try to make sure we have the time and like priorities to do it. And if you're at a company that has tons of meetings, like either stacking those meetings or being cognizant of a distraction free time. When I worked at Meta, they had an automatic focus time. So if you're coding like in Visual Studio Code, your Slack status, like the Messenger app internally, would flip to like a focus mode. And so when you got messages, it wouldn't even notify you, which is really nice because if you're in that deep 90-minute phase, you don't need to go manually say, don't message me. It would like kick in after a certain length in the IDE. And now you're like in deep focus mode. So I kind of like that approach. As long as you're getting stuff done, like, you know, who is to say otherwise? Yeah, totally. I have a newer thing I've done. this is more of like a 2025 thing. It's not specifically like a resolution, but just on that focus, on that topic of focus is basically like during the weekdays, I just have like my phone auto switch to like a, I think it's called like a work mode. You can name it whatever you want. But basically it's like it disables notifications. I think I did it from like 7 a.m. to like 6 p.m. or something like that. And I only get notifications like for my wife just, you know, cases emergency or something like that. And then like, my ring because like I want to see if someone's at my door and then like my credit card notifications because I want to see you know I'm buying something for someone to use my card they shouldn't be but like other than that I don't get any notifications like no slack no email you know it's just in it that's just on my phone you know so I still get my notifications like on my my work laptop obviously but just it kind of removes that like itch of being like oh who texted me or who you know put put a message out there who emailed me you know and it just helps a little bit you know I'm trying to to eliminate those kinds of distractions. So yeah, those focus modes, I think, are pretty good. So far, it's been a nice, nice development or incorporation in my routine. Yeah, I think the person who writes a lot of the content or like literature on that is Cal Newport. I was just looking them up, like deep work. And very interesting, if you're into that, I saw a tweet probably two weeks ago talking about monk mode. I was just looking it up while you're speaking. And it's, some startup, I think it was like Lindy AI, they essentially have this label that's like deep focus mode for an engineer in which they're working on something that's like extremely high priority for the next two weeks and they call it like monk mode or something. And what that means is this person like doesn't have respond to any messages, emails, attend any meetings. It's like they're in their own silo and manage their time on their own and just like get shit done, which I found kind of interesting. Like I can't tell you many times my life. I mean, I guess I work for myself. I was a little bit like that. But that was unique. in which, like, I've worked for a company and have the ultimate discretion on where to spend my time and, like, not have to deal with, like, the quote-unquote other stuff. But it sounds interesting. We'd love to give that a run through and see if, like, me being in full control of my time, not having to worry about other things, like, if that would be a huge productivity win or, like, you know, I kind of miss some of the other things or not enjoy it as I expect. It sounds like you're focused, but also super busy. Like, you know, taking a break every few hours, maybe it's not allowed in a monk mode. but kind of interesting. Yeah. Yeah, that's interesting naming convention, like monk mode. Yeah. What the monks would actually think about that, you know? Yeah. Are you adopting our name? Like, do you want to be a monk? Yeah. One thing while you're trying about notifications, I recently added push notifications to split my expenses, React Native app. And if you're familiar with push notifications for anyone listening, it's a huge pain in the ass. Apple makes it so difficult to set up from end to end to get push notifications delivered. But if you look at Split My Expenses, feature request board, so today I have like a feature request board where users can come in and comment on things. Notifications, I think, is within the top five upvoted by at least 30 or 40 users. So I spent at least six hours trying to scaffold out the infrastructure to, one, hook up with Apple systems, and two, on my code, essentially send a notification when an expense is created, a payment is created, a group is created, all these kind of state changes. within split my expenses. It's detecting those changes and then delivering a push notification to the relevant users with the right content. And I built this probably two weeks ago and I was dying for one of my friends to add me to an expense because I'm the only one in this system. It's kind of a closed door system, me to test it just for now. The mobile app is not out, but I have it on my phone. And just yesterday, my friend added me to an expense. I was like a dinner with six people or seven people and he added me and I tested it. I got the notification in. I was like, hell yeah, this is awesome because it was so much effort to get it to work. And I think it'll drive a lot of engagement because notifications of, you know, something is happening. I have email. It works. But the push notifications with a native app is like 10 times better. And so I struggled through it. I got the implementation done and was just like ready for a notification to get delivered to my device. And it finally happened yesterday. And the satisfaction was super high. I was like, this is going to be sick. I can't wait until I release this to the general public. but that feature alone push notifications is going to drive tons and tons of engagement of like, hey, go add your expenses. Hey, go settle up in the group. Like there's so many notifications. I think I have 12 or 13 right now that I've added. And it's probably it for a launch, but they're super impactful. And I'm really excited to like get those out the door. And once the app is finally out, then everyone will get them, of course. And I think it'll be a huge, huge win. So yeah, something I want to share because I worked on notifications at PayPal, enable notifications there for the business app and then doing it on my own for chatty butler and then finally doing it again for split my expenses and each time huge pain in the ass so much effort with certificates and going through apples kind of set up once you get it done it's like the code you never touch like you do it and you just hope to never ever touch that again that's cool you know that's cool to hear how much of a pain it is because I think you know for me just as a regular consumer I wouldn't know that they were such a pain to deal with you know, like it just seems like on the engineering side, it just seems like that wouldn't be, of all things that, you know, go into an application. It just seems like that would be just a quick add-on. But it's interesting to hear that there's you to get certificates and that Apple makes it kind of a chore to do. You know, you wouldn't have guessed that. Yeah. And then one other thing I added to SME recently was more frequent database backups. I won't go too deep into this, but previously it was doing once a day and I log into my Amazon account. and I back up to S3, which is their storage kind of like API or product. And I think I had, I don't know, maybe like two gigabytes total of backups. I think each backup right now is roughly like 50 gigabytes. Sorry, that doesn't make any sense. 50 megabytes, sorry, 50 megabytes for each backup. And I think I had like two gigabytes total. And the TLDR is like the backup strategy is once a day. and we retain like four weeks' worth and we start expiring old backups. And it popped in my head with more users now. I think I should backup more often. So this means if I had a critical bug that I deployed some code and something goofed up and I lost like some database date was corrupted, my next backup would be roughly like, you know, 12 to 24 hours ago. And I thought, you know, I don't make much money at all from SME, but the pain of like losing even free users or like losing their trust of, hey, I just added 14 expenses on my trip. I lost all my data because Brad had a bug, and therefore he rolled back to like the previous day, and I lost all that. So I upped it to actually every two hours, I think. And when I looked at my S3 recently, I think it went from like two gigabytes store to 12 gigabytes. I think the price was maybe, I don't know, like five cents to like 30 cents. And I think it'll roughly stay under 15 gigabytes. Like it's very, very cheap. My database will keep growing, so each backup will get more expensive. but for it to cost me like $5 a month, I'd have to have tons of data. And so I just realized at the end of the day, I'd rather have the security and safety versus like spending, you know, three cents versus like under $5. It's like, this is like a no-brainer. And my server is beefy enough to do backups more continuously without affecting, like, user experience. So I turn that on. And yeah, just a little bit more safer. I get Slack notifications on every backup. So every two hours, I get a ping on my phone, which I will mute because I used to get it once a day, which is like, oh, it work. And now I'm like, yeah, of course it worked. Every two hours, I'm getting spammed by my own software. So, yeah, those are things I've been working on recently. It's push notifications and more database backups. And that one's more for safety and security. But at the end of the day, that's like peace of mind is honestly super valuable. Yeah. Well, and it's those things that, you know, business continuity plans, like I've worked at places. There's one place in particular that I've worked that really values that. And like business continuity plan, at least how I view it is like, you know, what happens when, like, you know, crap hits the fan like you know like and it goes through any kind of different scenario like if our servers go down like what the what the backup like if a facility is attacked like what the backup like if we lose all these records and like just having a plan and like a series of steps that you think you going to do I think it a very mature thing for like a company to do you know and you taking care ultimately of your customers that way And, yeah, it might not be, you know, a hot new feature. But, you know, I think people want to know that their data is being held in good hands. And storage space is so cheap. You know what I mean? It is. Yeah. Why not just do that, you know, and just it doesn't. It's so, like, cost-efficient to just add a couple more per day or whatever. Yeah, it was super cheap. Yeah, one thing, one quick thing on the notifications, then we can move on off that topic. But the, you know, we both last episode talked about our duolingo streaks and what we're doing. Oh, yeah. I have to say, hands down, the most passive-aggressive notification system. I agree. Yeah, they are. I mean, they kind of meme into it now, you know? Like, there's a whole, like, they put out, like, albums on Spotify, where It's like, and there's like songs where it's like, they're joking about how like, if you don't, you know, do your Korean lesson, they're going to break your knees, you know. They lean into like how, you know, ridiculous their notifications are. But yeah, I just thought. Yeah, I get those notifications too. Like, if I don't complete the lesson in the day, it's like, hey, your time is running out. Hey, are you sure? Sad face. Sad face. I'm like, I would love to do this for my own app, but I don't exist in a category to be used every day. But their playbook for notification delivery and like almost self-gilt is. is incredible. And you can take a look at their stock. Like if you look at Duolingo stock, they are doing quite well because they've mastered monetization and subscriptions. I think notifications and like user experience feelings where you want to do it, they know you want to do it, but you might not have time, but they just try to squeeze in with a notification at the right time. And I'm sure they have tons of engineers working on machine learning and AI to figure out notification delivery time. Because that's like a big sentiment is like if Bradley does the lesson at 8 p.m. every day, you know, we're going to send notification at 730 and say, hey, it's probably time. You know, they're not sending it randomly. They're thinking deeply about that. But yeah, it's an interesting area and very valuable. Yeah, I even have the tile. You know, you can have like tiles on your iPhone. Yeah, the widget, yeah. So I have the little duolingo widget. And if you, I notice, like, it changes throughout the day. So, like, in the morning, it has, like, a time to practice. But then if last night, I didn't do it to like 1130 at night. And like when I went to it, the widget was like, an angry, like, bird. You know, it's like, an angry duo bird. I just was like, man, this is crazy. They change the app icon, too. I think it's the same thing where they really pull all the levers that Apple tries or like it allows you. And they do it in a way that is like very growth hacky, but kind of interesting too. It's not like over the top. I'm going to uninstall. It's like they're pushing you in a way that makes you feel comfortable. Yeah, yeah. And it's in good fun too. Yeah, yeah. People that hopefully don't take it too seriously. But yeah. Yeah, that's pretty cool. And on your work, I know you texted me about, you know, kind of starting on the chat operator kind of phone call thing we had talked about last time. Any updates to share on that? Yeah. Wow, I just looked at Duolingo stock and yeah, you're right. They are killing. Yeah, they're doing quite well. Yeah, good for them. That's cool. I think they're based out of Pittsburgh, if I remember correctly, which is not a very common tech hub. Anywho, yeah, so originally I built that with Flask, you know, and I think I had talked about how I was using, you know, chat and like Anthrop or Claude Sonnet to, you know, give me some of the code. And then so once I kind of had it to a place where I wanted to make it a bit more, I don't say legit, because it's still very much just like an idea, but just a bit more robust and familiar. I switch it over to Django. And so what I had was trying to do, and this is a very classic, you know, Bennettism. So just bear with me as I walk through this. What I was trying to do is I said, okay, I want the back end to be something I know because you need to have some familiarity. Otherwise, you're just not going to be able to like, you know, you are an engineer. So like you don't have or maybe you might have something similar if you go into a whole different like language that you're not familiar with. But like, you know, you can't, you can try new things. But like if you don't know kind of what's going on, like you're not going to be able to get very far. And I don't want to spend a much time learning. So I was like, I'm going to have my back end be Django. But then this is where I made this the same old mistake I usually make is I said, okay, well, I want, I was a little inspired by you. I said, I want to be able to have like a mobile app too, like, you know, a nice, like native mobile app. So I was like, well, I might as well just, you know, Brad's talking about React Native. I don't need to know React. Like I can just see how easy it is for chat GPT to like give me the code base to set this up. So long story short, I went with a, the Django Rest framework is like a Django library, which is basically turned. your database into like an API. So basically it's like a database. You have like serializers and then you can just make like API calls with like a front end like React using like Axios. And so I had the back end all set up. No problem. Like I know Django decently well. And then I get to react. And because then my thought was like if I just make it into React, then it'll be really easy to go React native. And like and then it's basically the same code base. I asked like chat GPT that question. So I don't know if it was actually true or not, but it basically was like, hey, if I just make something in React, like, is it really easy to make it like a mobile app with React Native? And it was like, oh, yeah, it's definitely easy. Like, blah, blah, blah. It gave me like an AI answer. I was like, okay. So we'll try that. It could not, just could not work. You know, I had the front end folder, had like the, you know, the component set up with like the router and just could not get it to work. And I kind of went into it, like knowing I'm like, see how well this works. I wanted to kind of like, I wasn't like pinning all my hopes on that, but, you know, I was like, let me see how far it would get. And I think two things, and I'll pause there, because I know you probably have a couple comments, but the two things that I noticed in just doing that exercise alone is one, this probably prompts engineering things that I'm not doing that I could do better for sure. Mm-hmm. And I think my takeaway is like to try and narrow it into smaller steps because it kind of would not just give me the whole thing. And it's like, well, that doesn't really, it just doesn't plug and play like that. But then, two, it was not giving me like any like tests. And like I, you know, test driven development or whatever, I think is like a philosophy that people, you know, really stick to or they don't really think that's that important. But like for me, again, as like a non-technical person, I really wanted to see like, okay, does this part that we just just so far? Like, does this part work? Because I don't want to get all the way down this path and have something not work that we did, you know, 10 minutes ago. You know what I mean? Like I want to kind of like sequentially, okay, you're telling me. I need to put this in this folder and then hook up this router over here. If I just do that, does it work? Because I felt like it was trying to get me all the way to the end results. And then something would go wrong. But it wasn't really clear where things broke. So that was like my experience. But I'll pause there because there is more. But what do you think so far? Like what is that? Yeah. My first reaction is wow, because if anybody knows Ben, the JavaScript and him do not work very well together. And to jump ship to React, the leading JavaScript Web framework and React Native, It's a big step. And exactly as you mentioned, being familiar with the one thing is really powerful. For me, you're using Django and Python. I use Larval and PHP. So that gets me, that domain is satisfied and easy to work with. The front end, I've historically done HTML and CSS raw. And then I switch to view because Larval kind of picked that up and pushed it in their ecosystem. React has now pretty much been the dominant force the past few years. So so much stuff is built on top of React, including React Native, which is a huge win if you need it. it. So I think my first impression is I'm proud that you went that route. However, I know it's bumpy and bruisey for someone who's not familiar with the tooling because I think the comments that I hear from you historically was it's not that easy to build a website given all the amount of tools that need to be scaffolded for you to become productive in the beginning. Because when you're working on a Django web app, it's very simple. Like you install a package, boot up your server and your writing code. React, React Native. All these things are. are multi-layered systems that all run locally, and their configuration can be extremely complex and time-consuming. And just like, if you're not familiar with, it's a huge pain. So I think my first reaction is I applaud you for going down that route. And my second reaction is exactly as you mentioned with the querying AI in terms of fine-grained updates versus tackling a larger project. I think it's exactly right. I've also been toiling with how to approach AI in ways that makes sense for building my app. like on nights and weekends. And one approach to that I've been trying to do is to distill the context of my project into a generic prompt, which gives like the high level context. And then for the problem at hand, I describe another section to the AI that says, this is the task I want to do in this system. And so I think a few weeks ago, I won't go too deep into it, but I wrote maybe, I don't know, 900 words describing split my expenses is the bill splitting out. You can think of it as split-wise. And I only say that because split-wise is much more index in the AI world, not necessarily that I'm an exact copy, but more they understand what that means. Then I describe my database schema, so whatever my main tables look like in MySQL. I give it sample of test data rows. So on my local machine, I can spin up the database and like seed fake data into it. Because that gives it a feeling of what's actually in there. You can say an expense has an amount and that's an integer. But usually for expenses, it's like, you know, below $10,000 that people are splitting. So it gives you that extra behavior. So I described like the overall Larval, PHP, Vue.js. These are technologies. This is my database schema. These are the main features. And then after that I have like, this is the feature that I'm working on. Here's the code that I have. Please help improve this given the system constraints. So I'd recommend that for you. I think that would be a good way to approach it. I think the prompt engineering stuff is actually extremely difficult. And historically, people have said it's going away. But using models like 01 pro, it almost feels like it's even more important. to say the right things. Like, they're smart as hell, but if you give anyone a generic task of, like, make me a React Native app, even the top engineer at any company will be like, well, there's so many different pieces in that. I don't know how to succeed given, you know, give me enough info. So I think those are my thoughts. And I think React to React Native, I'll touch on that last, is useful. But I don't think it's as easy as it might sound from just the naming of those two projects. Like, React, you're building HTML and CSS. React Native, you're building, you're using different components. So in React, you'd use a div or a P tag or a span tag. And React Native are using like a, you know, a text or a label and like an image tag. They're just different and they have different constraints and different libraries. Where I see React Native shine is it gives you an opportunity to write shared code for 80%. So like maybe your data management layer is shared. But that other 20% of making good UI on both iOS and Android, it's still really freaking hard. Like it's not easy to ship something that feels good. So I think in your case, you would definitely win with shared data processing, but the UI is still, like, going to be really challenging. Yeah. And I think, you know, one thing that I noticed it was doing was it was, like, adding all these features. Like, I would tell it, like, this is what I'm trying to make. And it would, like, go ahead and start, like, adding a bunch of stuff. Again, because, again, it kind of went to try and get all the way to the end solution when it's like, in my experience, just building little projects here and there. It's like I you kind of and I don't know if you'd feel the same way, but like you kind of like work on an individual piece first. Okay. I'm going to get like the login. Yeah. Like I'm going to get authentication done here. And I'm going to go to, you know, user management over here. And like, but it would just try and get everything done in one and then like add things I didn't really want or like wasn't ready to add. I was like I just want to get like an MVP kind of idea going. But it was like, let's also have it be able to do these things too. So I had like when my prompt is doing, I was like stop like telling me what I need to do. I'm just going to like just only do this, you know? Yeah. So, but it goes to show, you know, I think we've talked a lot and people talk a lot about, you know, like certain things going away because of AI and like, you know, certainly there is potential for that to happen. But like, it's just not there because, you know, even like, you know, I'm pretty familiar with Django. And there's a really good authentication library that people just use. Like it's very like standard. And it didn't push me to use that. It was like, oh, let's handle authentication. And it was a bit more like bare bones. And I was like, why don't you just use the camera? I think it's called Django All Off, like All-Off, like, A-L-Off. That's the good one. Yeah, yeah. I mean, last I checked. And that's what I've used into other projects. And it's really easy to set up. It has like built-in views and all that. So, you know, log in, reset password, confirm email, like all that kind of out-of-the-box stuff. And it didn't, it didn't use that. And I was like, why wouldn't you just use that? And it was like, oh, you're right. Let's use that. Just like, okay. Like, why didn't you start with that? You know, so. But yeah, so it's been, it's been good. I haven't been able to get, you know, to kind of come back to the idea. I haven't been able to get back into it since the last thing I ended up happening is I just said, scrap React, I don't care. Like, it was just cool thought. But like, I'm just going to just try and get like MVP up with. Yeah, yeah, exactly with Django. And just get the functionality to work. So I was spending a bit of time with like Google Texas Speech Library. And that's where I like, I stopped. and I got busy with other things. So when I come back to it, I got to put on the bingo card getting Ben to use Larravelle in 2025 because that would be a huge win. Then we can just hack on stuff together nights and weekends and I can do a pull request into his repo on GitHub and fix some issues. Like, Django, I can do it, but I'm not at home. And I think since I'm an engineer by trade, if we got him, you know, if you're listening, maybe reach out on LinkedIn and give him a thumbs up to Larerville. But if we got him on Larval, we could, you know, jam on things together. I've been pushing it for a while, but yet to be his framework of choice. They don't play nice on Windows. I don't care what people say. It doesn't play nice on Windows. So I know that there's like Larravelle Vapor or whatever it's called or Larval Sale with like Docker. Well, yeah, sales docker. I get it, but like it don't play nice. I tried. Okay, here's my take. I'm going out to see Ben soon. Maybe I can go to his computer, get him set up with the, you know, me a Laravel developer experience on Windows that I've heard exists, although I can't confirm. After that, maybe in the next podcast episode, he'll change his mind. Or we can double down that it's not good on Windows, and I can reach out to Taylor and say, you know, clean this up because I can't get my brother on it. We'll see. We'll see. I have, they say, oh, use the Windows subsystem for Linux. Oh, yeah, WSL, yeah, yeah, yeah. No? Not for me. Not for me. Not for me. It's on it for me. Again. I did tell Ben, like he. He told me he was working on this. He's like, hey, you know, I'm like starting to commit to my repo. And the first thing I say is Larval and a MacBook. I say that's your starting point. But he didn't like that one. I'm a PC guy. I got my laptop right here. I'm on my PC right now. It's good for gaming. It's good for gaming. Well, it's good for business. It's good for business. Mac, it's just, look, I have a Mac for my 9 to 5 for now. for now. Yeah. And to tie that one off, I think exactly as you mentioned, for you as someone who is technical, like you've deployed Django apps to the, the worldwide web, you've done with, you know, you've dealt with libraries, you've done Python Excel. Not to say your skills aren't good, but even with the help of AI, like things are still hard. I think there's this general consensus in the software engineering space, especially on Twitter. Every week, there's a new AI model that comes out that, you know, does pretty damn well and maybe it's pretty cheap. like software engineering is over. You know, I went to college and got a computer science degree. I've worked for the past, you know, seven to eight years building, you know, mobile apps and websites. I don't think we're there yet. Like, yes, these models are getting a lot better. But you almost think, like, this is a good litmus test for, you know, Ben, who's someone who is technical, can he build like a React Native mobile app without running into any hurdles? Because I think the day that software engineers are gone or the day that you, you can get to exactly where you want to go without running into hurdle after hurdle. And I think React and the complexity of building a solid web app with good foundations is already hard. And then to make it even better is like way harder. Like you could probably get off the ground spending a ton of time prompt engineering to get the right solution. But I almost guarantee you whatever comes out of that, you'll be like, this is good but not great. Like there's the existence of apps in the good space on the app store and websites that exist. and it's like an MVP at shippable, but those ones like Airbnb, the Netflix app, like Stripe, the ones that really wow you, like no AI right now can do that. And the second that that is at a click of a button, I'd be a little bit scared, but even then, tying pieces together, like you mentioned where AI will just go down rabbit holes and doesn't think holistically about the big picture, once we get those two components where it's able to execute at a high level and a low level and tie it back, and it's able to innovate in ways that it can mimic modern UI design and be transformative to create new experiences that we haven seen before that are better That a challenging point but we far away from that And I think this is a good understanding that yes it helpful but you need to guide it It's like an intern or like, you know, a new college grad. They're skilled, but they need a little bit of direction. I think that's where AI exists today. Yeah. And it's also, too, like, at least in the stage right now, it's like, you might tell, like, someone, like, I guess the idea is that like people would just be able to write their own software, just because they want something, you know, just any random person. But it's also like, you don't know how to do that. And sure, like, maybe the AI might, but like there's so many questions as you go about building something. Like, again, just with this like call thing, it's like, I want something to call me at certain times and talk to me. It's like, okay, like, what are you going to do? You're going to use Twilio. Yeah. Like, how, you know, like, have you thought about what that's going to cost you? You know, Google text to speech. Like, there's so many different design decisions that the AI would have to decipher and choose from. And, you know, one, we've already talked about how. it doesn't like take easy wins like the all off library like it doesn't and then too like I saw i think ben Affleck the actor was talking about AI like taking over movies like and making its own movies like people are only going to see movies that are made by robots and not be you know the Steven Spielberg arrows are kind of over and he had a good point and you could tell he thought a lot about this and his response was he basically i'm paraphrasing but he was like art and like building something is like also knowing when to stop, you know, I'm like, and knowing that something's finished. And he's like, there's, he was, he was much more adamant, but he was just like, AI is not going to know when to stop. Like to the, yeah, step back from like the, you know, Sistine Chapel and say, okay, this is enough. Like, this is it. Like when DaVinci was painting the Mona Lisa, when he stepped back and was like, okay, I'm done. Like, he's like, AI is not going to be able to do that. And like, you know, it's so that, that one point, I think, I think it's a good point because it's just like, there's so many different decisions that go into it. It's like people, one, they don't really know all that goes on when they just say, oh, I want a software that does X. You know, there's a bunch of different design decisions that go into place. And then two, you know, I think a little bit of like Steve Jobs where I think he, I don't know if this is like actually should be attributed to him, but I feel like he would say that like people don't know what they want. Like you need to show them what they want, you know, and it's like. And so yeah, people might want software that does something and they might just think of it and maybe one day they'll be able to just say oh I want you know the software that does this and they can just write it themselves but like people you know it's just I just don't think it's going to work like that I think they're going to just rather have something be built for them and and not have to worry about like making it themselves yeah I've seen a tweet recently that crystallizes that and it was saying yes you can get software written for you by AI and it might work but maintaining that running that on a server all that complexity and and like uptime and making sure it doesn't break and then adding features to it over time and choosing the right technologies like you mentioned, that is like almost the harder part. Like, yes, anyone can get, I can ask AI to write me Python code that transforms like an Excel sheet or, you know, goes to large extents to manipulate local data, but what if you want large scale data? What if you need privacy? Maybe you could go pretty far with that, but like it's hard to get it the last mile of, you know, having it continuously running and be supported and not break. I think when you pay for things like a task tracking system, like linear, Gira, all these kind of products that seem easy, like a to-do list online. Yeah, anyone could ask Chad GPT or Claude or any of these models to make a, you know, a to-do list app locally. But like having all these integrations and other features and making sure it doesn't break, making sure you can access it, you know, outside your home network. Like, good luck with getting AI to do all that for you. Like, it's just not there yet. So I think we're getting close. there's new models that are popping up and we'll probably talk more about them, but it's not at a point yet where I'm concerned. I think there's been so much hysteria online that it's almost like a little bit negative and just like annoying, almost at a certain point where every new model that pops up, we say this is it. And pack your bags, we're done here. And it's just not like that. And so I can tell you here, at least in January 2025, I'm not concerned. I'm more than ever excited about the progress. And I think it's only going to keep getting better. and I'm trying to use these tools as much as I can. I don't want to be left behind. I try to lean on it pretty heavily. It's almost kind of like I use Google Maps. When I am navigating in my own neighborhood, I know where things are, but I like seeing a time. I like seeing a route. I like seeing the optimal route. Maybe it's not the same route I used to take. And it's like, yet do I rely on Google Maps? Sure. Like, could I have a better mental map? Sure. And I think the same thing exists for AI where I can do all these things manually, but if there's a workflow for AI that I can give it to it and it can do a 90% job, 95% job, I'm going to lean on that. Like, I'm not going to go do it manually anymore because I think it's only going to get better and better and handle more complicated tasks. And my job, at least as an engineer, building these, like, local workflows, is getting at the data it needs in that context. And exactly as you describe it were, you need to really have a high-level plan in place. So you're directing. It's not innovating in ways that you're not able to explain. You're able to, like, tell you you want innovation in this area. and this is your long-term direction and just solve one thing for me. So that's what I've been trying to do recently is, one, use the latest models, two, distill workflows down to AI in which I get the maximum value out of it by reducing the amount of manual labor I do, but it's well-defined in an area that's not like super complicated. Again, a mini update from last podcast is I did subscribe to O1 Pro, which is $200 a month. Again, we talked, every AI tool right now is $20. Claude, Open AI, to chat GPT Plus, I think is 20. Google Gemini is 20. Everything you can imagine is $20 to have the user experience to chat something with high intelligence. I went down the rabbit hole for O1 Pro thinking, what can it do for me and how can I leverage like the quote unquote smartest model from Open AI? Because when you think about it, I feel like I'm solving decently hard problems with my expenses. Like there's areas where I need to really think, like I mentioned earlier in the podcast, I need four to five hours to work on something. And the current example that I'm doing is currency conversion. Just complicated to get right. I want it to be 100% right. And I haven't done it yet, but I need to throw O1 Pro at it. I've thrown O1 Pro at different features. Like, for example, recently I upgraded some of my activity log code. And activity log for people who aren't familiar with SME, is it just tracks what happens in the system so that if your friend adds five expenses and then deletes three of them and you get these notifications, you can kind of see what happened over time. And I needed AI to add more UI for a new activity log type. And what I did is I gave 01 Pro the context of my file. So like my view front end code, my Larifel backend code. And I said, this is my system. Again, using that prompt that I described earlier, outlining the full core tables. Then I defined the feature that I wanted. And it did pretty well. I use a tool called repo prompt, which I'll go into later. But essentially, I use repo prompt with O1 Pro. and the final result wasn't exactly one shot. And one shot is kind of the essence of you give it a query. It produces code. And that code works flawlessly without any errors, whether that be syntax or logic errors. I think when I applied my code from 01 Pro, I think it had like an import error or two and then like one or two logic errors. Otherwise, it's pretty good. I wouldn't say it's necessarily worth it. But they've also came out with operators. I don't know if you've seen that, but it's super cool. Open A I ship both operators. and scheduled tasks recently, which we should touch on. But overall, my 01 Pro reporting is it's good. I need to try it a little bit more, push it to its limits. I'm currently paying $200 and my timeline is running out for my 200. So probably the next week or so, I'm just going to throw it at the wall and see what sticks. Yeah. Yeah, if I had that subscription and like the operator and scheduled tasks thing is huge. I mean like just because you want people want to be able to have something happen like on their behalf without them actually doing it, you know? Like the whole like make money while you sleep thing, right? Like that's why that's so repeated. And so, but the fact that something can occur without you like physically manning it is, I think a huge feature as a whole that people are really looking forward to. So like that, that definitely, you know, makes the subscription worthwhile or, you know, at least an interesting endeavor. But if at that price, I'd be like, I need to use this a lot to get my money's worth. Yeah. You know, like making sure I get like to spend a lot of time in it to justify that. but I'm sorry if I miss this, but is the operator only part of the premium subscription? Is that kind of? Yeah. So operator is on their pro tier, which is the $200. The plus tier has only 01, but not O1 Pro. So O1 Pro is like their, you know, highest intelligence takes a while. It's the thinking AI model. And I think scheduled tasks comes out for everybody. So I'm not super sure. I often take a look at it. but I think it's for all chat GPT users. Maybe there's a limit on like how many tasks you can schedule in the frequency. But I think overall operators is out now only available for pro users because I think that's limited and new and that's kind of their way to roll it out in like a safe manner. And scheduled tasks is you can schedule anything to happen on the chat GPT platform and it'll notify you and like give you information that you requested. And I think that combined with like web search, decent intelligence. It's interesting. I haven't found a ton of cool use cases yet. But on Twitter, I'm kind of like waiting for these, you know, people who make these threads on AI to come up with some cool use cases. But I did try Operator. And what I tried it to do is my first query was give me the best bill splitting web app. So as someone who makes one in the space, I thought, you know, people are typing this into Google. Let's see how it evaluates bill splitting web apps today. And what it did is it did that exact query, like best bill splitting web apps. I think it clicked on the top three links and it scanned the web page. So for those unfamiliar with operator, it's essentially a chat interface on your left side and a web browser on your right side. And this is within your own web browser. So the web browser that you see is a video that's streamed in of OpenAI controlling a web browser. And so you can see it go in and search Google. You can see it scroll through the web page and it opens up multiple tabs. So it's just like a normal Google Chrome browser. And at the end of looking at like three articles, it was like, hey, splitwise is looking like the best, you know, web tool. And in my head, I was just crying. I was like, I paid $200 and this thing told me, it's not even me. So I then asked you. I said, yeah, I said, what about split my expenses? I heard that one's up and coming. And it did more Google searches. It found my product hunt launch, I think, from 2023 and said, hey, SME looks pretty good and users like it. And I thought, okay, like, what else can I do? Like, that's just reading Google. It's not that hard. So then I asked it, can you please sign up on split my expenses, like create an account? And I thought, like, what is it going to do? Like, does it have access to, like, an email account? or whatnot. So it first off told me I am not allowed to sign up to any services. I thought, okay, you know, Open AI clearly does their due diligence on like, I don't know, maybe malicious behavior. One can say you could use operator to do a bunch of sketchy stuff and like deem it was Open AI. So it first told me that and I said, okay, like, can you at least walk me through it? And so what it did is it went to the sign up page and was like, this is where you'd fill out your information. I thought, that's kind of useless. But what I looked on their docs is you can actually, take over the web browser. So what this means is, say if you have a complicated workflow where you log into Google Drive, then you log into YouTube. For example, we upload our podcast, we use four or five different websites. What operator can do is you can log into each of your accounts on its own Chrome web browser that's remote because you can take control of it. After that, you can tell OpenAI what you just did. So when you take control, and then you release control, it says, what did you just do? And you describe, I went to transistor. our podcasting site. I went to Riverside FM, our podcasting site, logged into my admin accounts. Then you could describe a workflow in which we use multiple tools to coordinate and clean up our podcast and upload it. And it seems pretty cool because I think it can resume, like on a new operator chat, it can pull out your already logged in accounts. So you can imagine this is actually how you get the maximum value out of it. I really got little to no value of asking it to Google search for me or try to sign up to my site. But what I want to try next is giving it my credentials into various systems that aren't super sensitive. Like, I don't think I'd give it my bank account login. I'm pretty sure I'd probably block it. But I think I've seen some sites recently that if you try to ask you to search YouTube, for example, it's blocked unless you log in. And I think opening I worked with these websites to say, like, this is the operator web browser. So, like, you can identify it and see it on YouTube's or like Reddit side. I think Reddit is also login only so that people need to be authenticated in these systems. but overall super cool. I'll probably report back by the next pod of giving my credentials to a deeper system and then seeing if operator can do something meaningful out of it. Our friend Justin Anderson has been posting on LinkedIn about doing a little bit of that. I think he logged in with his Google Drive and just made some mock spreadsheets and is doing some coordination between different spreadsheets, which is super cool. Yeah, yeah. And he's in the accounting industry like myself. So that's cool. Yeah, he's got some good. content out there and it's it's I like it because it's um it looks super practical like it's not just like you know I feel like a lot of um you know I guess I don't say like YouTubers but like a lot of people just take like the textbook use case and they all just do the same use case of like a to do list like everyone has a to do list you know tutorial like if they're like talking about React or whatever but the Justin's content seems very like specifically curated to accounting of like hey, I'm taking, I think you had like an expense, like receipt categorization that took it from a PDF and put it on a Google sheet and then categorized it. Like, just very useful and practical, which I think has, you know, is what distinks content from, you know, generic content, you know. So, yeah, shout out to him. It's good stuff. Yeah, it's cool seeing people make content with AI because there's so many different use cases, like the ones I think of are engineering, but for the accounting side, like seeing your guys' point of view with the tools at hand. Like, I wouldn't even think of some of that. Because my day to day is not doing some of that or overseeing some of that. But my day is doing other stuff like in Python or iOS or Web. And I'm like, how can I speed up my workflow? So really cool to see Justin doing that. One thing. One thing that we're going to clip right here for the audience here. Brad just said is he says he's doing something in Python. So why would he need to convince me to go to Larravel when Django was in Python and he knows Python? There it is. I'll fess up. There are plenty of things that I automate in my life that I use Python. and that is because Python is widely understood and it's very easy to run. You don't write any local PHP scripts for automation and use Python. So Python has its place. I just, you know, Larravelle and PHP are just king in, you know, web development. Ruby on Rails is great too. You got Larval and you got Roar. Django is definitely a second class citizen, although it does the job at times. Yeah. It's interesting. I almost done a ton of time on this because I know we got lots of other cool stuff to chat about. But the fact that you said that, you know, Ruby on Rails, Larval, you know, Python being the most widely used language, it's where they don't have, like, a real dominant, like, web framework. You know what I mean? Like, Flask is there. I think that's probably, I don't know the stats, but I guess, like, Flask or Django is probably the top one. And then I think fast API is, like, good just for getting, like, APIs up, obviously. But they don't have, like, a Larvaeval. They don't have a Ruby on Rails. They don't have, I'm just trying to think of other, like, web frameworks, like a, like a React. with no. Yeah, Next.js or any of those. Yeah, yeah, it's interesting. But anyhow, yeah. Yeah, someone will make it. If anything, they have a good example to follow with Ruby on Rails and Larravel. So it's just a matter of time if people care about it. But I mostly see Python use for local scripts on automation and then machine learning. So I think that's kind of where everyone's focus is, at least in that area. Yeah, yeah, for sure. That's cool. Cool. One thing that we had on the list here that I really wanted to kind of hear. you speak about was Olamma plus open web UI because I had, I think we had talked about it, either it was on the podcast or just, I think I maybe sent you a video a couple months ago about open web UI, but I never was able to really like get into it just, you know, for lack of time. But what are you seeing now with that, with, you know, some of the new developments that have come out recently? Yeah, so DeepSeek R1 is the newest AI model. I came out of China. And I'll be honest, I haven't read like the research paper or anything, but the TLDR is it has on-par performance and lots of various categories like coding, mathematics, logic, arithmetic, all that sort of thing with Open AIs O1 model. So again, I pay for O1 Pro, which is technically better, but DeepSeek R1 is on par with Open AIs O1 model. And I think it's 96% cheaper. So you're getting a high intelligence at, you know, a fraction of the cost. And it's open source. So this means I can download this model and run it locally on my own MacBook. But again for people who are familiar with you know AI machine learning models the model that they are running on a large large GPU is much more powerful than the model that I can run on my local computer because you need I think their biggest model is like 405 billion parameters and like, I don't know what the exact translation is, but you can imagine you need lots and lots of RAM, gigabytes of RAM to run that. So maybe like 400, 500 gigabytes of RAM to run it. I don't know if that's right. Don't quote me on that. But at the end of the day, this new model came out from this company in China. It's extremely powerful nonpower with open-AIS closed source model and it's 96% cheaper. So this is a huge deal. And then I wanted to run this model and my Twitter if I pull in my Twitter right now as we're speaking, probably six out of 10 tweets and my top 10 and my 4U page would just people saying, oh my God, deep seek R1, incredible, research team, incredible, the vibes. So if you're not in the AI space, they talk a lot of about vibes, which, you know, is one of those things that's hard to put your finger on, but it does what you need to do in the right way. And it's scale and compute and all these factors make it really, really attractive. So meta comes out with Lama, which is their open source kind of behemoth of a model. And they're kind of laughing online right now because Deep Seek R1, like leapfrogs, Lama, because Lama was pretty good and costs a lot of money for meta. And Deep Seek R1, the biggest thing that they came out with was a highly intelligent model that's cheap and their training they supposedly trained it i think on like 50 million dollars i don't know if that's the exact price but it was extremely cheap so coming out with frontier model intelligence and the training cost of building the model is 50 million and it's cheap to run is like a huge combination of wins and it's open source it's like an absolute game changer there's tons of people on twitter that are saying this is like a pinnacle of engineering and research and now that this is open source a bunch of other new stuff will come out in three to six months that are going to like make this even better. So that was a long tangent, but TLDR is deep seek R1 brought me into, how do I run this at home? Because this is a really good model. So I usually use Claude Sonnet, which is the best coding model. Deep Seek R1 is supposedly better in some areas. So I was like, all right, I'm going to use O-Lama. And so we talked about on the podcast before, but O-Lama is a way to essentially download models and run them on your local computer in an optimized way. And the DeepSeek R1 original model is like 405 billion parameters. And that takes up 400 gigs of RAM. What I'm running at home is either their, I think, 30 billion parameter model, which is like roughly 30 gigabytes of RAM, or I think they have a 60 gigabyte model, which is roughly 60 billion parameters, which is roughly 60 gigbytes of RAM. So I use Olamma to download the model. So you say like Olamma run and then the model name, which I think is like deep seat dash R1 colon, like the size, which is 30 or 60 or 7 or 405. That's great. Olama gives you a command line interface. means on your terminal on your Mac, you can type in anything. It's like good, but not great. Like, you know, working with chat GPT or any modern chat interface has like code formatting for text. You can like copy blocks the code that it outputs all these things. So I was like, how do I get a better interface? And I was looking around, especially on my Twitter right now, tons of people are talking about it. You even mentioned it. Open web UI. It's honestly a clone of chat GPT's interface, but open source. So it's like a normal chat interface, you get your chat bubble. It's probably written in React. It allows you to upload images, like, anything in the modern day for a chat interface. But it's all free and open source. It's incredible. And it hooks into O-Lama. So open web UI is just a website you run on local hosts, like local host port 3,000. It then sends your query to Olamas's running model on your computer. So you get a fully offline and local feedback loop with a high intelligence model. And it was just mind-blowing. Like I couldn't believe that I could get this set up within five minutes. And that five minutes probably includes like the long download time for the deep seek model. But literally what you have to do is install Olamma, run one command to download the deep seek R1 30 billion or 70 billion parameter model. That is like a 40 gigabyte to 20 gigabyte download. Then you run one command which uses Docker, which Docker is complicated. But if you go to open web UI's read me on GitHub, there's a hey, if you use open web UI and want to use OLama and OLama's already running our computer, copying paste this one Docker command. It'll then, you know, run the open web UI locally and connect to OLAM. And I thought, this is going to be like a huge pain in the ass, but, you know, I'm going to go down this route. Literally one command who was up and running, went to local host, it automatically detected that I have the deep seek R1 model installed. And I was like off to the races with a extremely high functional chat interface. And it's all running locally. So I could give it like literally my personal information. I don't really know on the full security. I'm like 99.99% sure it's safe and there's no like leak in open web UI's code or Olamas code or the model that it's running. Like those are the three pieces of software that coordinate it all. But I can literally give it like my finances for Bucksby or like any other personal information and ask for advice. Or if I'm chatting with Claude, chatting with chat GPT, yes, they don't like always train on your data but I wouldn't tell it like my social security number, for example. Not that I am, but that just gives you the context of which there's really no internet connection that's going out there. And so, yeah, it was a very interesting setup and felt extremely high class. And I think Olamma and Open WebUI are a good foundation. And DeepSeek makes it extremely valuable to plug into that system and have high intelligence locally on your computer. So long story short, an extremely cool setup. I think everyone should try it. If you have a MacBook, if you have 64 gigs of RAM, I know it's kind of a lot. But try it out and let me know what you think. run like the 30 billion parameter model and it's pretty damn good i was i was blown away yeah i saw it um a couple months ago from a youtube channel called network chuck and he makes pretty entertaining videos he's definitely um any he disclosed as he like promotes you know products obviously it's making a living right so um but he had it the way he set it up and i actually i was like that's a great idea is he you know used llama like so that's model, you know, so this was pre-deep Seek, this couple months ago. So he was using a llama, and he basically, his use was like, for my kids when they're doing their homework, like, I want them to be able to ask, like, for help, like tutoring, essentially, but I want to be, like, to control, like, what they can, like, get information from, you know, obviously. Just, of course, as parents, you want to make sure your kids are getting the right information, not seeing things they shouldn't. And so he had to set it up around that, which was pretty cool. And then I think he was using, I have to go back and look at the video, but I think he was using like maybe one of those NAS. I think it's like a local server. I can't remember exactly, but I have to go back and look at that video now because I'm curious what he ended up doing with it. But yeah, I mean, it was cool. You could like drop down, pick which model you want to use. I think you could even use more than one in certain cases. But then he like, you know, using Lama, he like kind of configured a model file that was like specifically instructed to like not just give the kids the answer. like, you know, explain, explain something like lead them to it, you know, and kind of act as like a teacher more than just, yeah, exactly. So, you know, because I remember using Wolfram Alpha back in college. Oh, yeah. And it would give you like the thought, the theory behind, like, but I was like, I just want the answer. But, you know, so thanks for all the theory, but I just want the answer. So, yeah, we'll link that video because it's a good video. He does a good job with his videos and, like, editing and all that. And it kind of gives people a visual to like open web UI. they don't want to check it out themselves because it seems like a pretty cool tool. Yeah, it has all the bells and whistles. It has text to speech, speech to text, live conversations. So chat, you can with that in December last year. Like, it even has a rag, retrieval augmented generation. So you can connect it with your Google Drive and it can pull in documents. Like, it's honestly pretty incredible. And I haven't configured it to my likings just yet. Like you mentioned with controlling the AI's behavior, you can put like a global prompt in and say, like, if I'm querying Lama, include this prompt, and he's probably writing, like, you know, you're a teacher, you're teaching kids under the age of blah. Like, they are very curious, but don't give them the answer. Like, almost the prompt engineering behind the hood, and then the kids will just access the open web UI interface, and it looks like Chad GPT, honestly exactly the same, but you'd have to do a bit more effort to, like, expose it to your local network, like you mentioned, setting up some at-home server and such. Yeah, super cool. I imagine it's only going to get a lot, lot better as we move forward. like we have the foundation of the web UI plus a model runner. Now we're just getting better models to come in and feed them. So it's like a very, very exciting time. And I can't wait to see what comes out of Deepseek's impressive model release that, you know, I think people are really, really excited right now in the past, like just a few days that came out. So if you're not aware, definitely check it out. It literally takes five minutes. We have a MacBook to download Alama, download Open WebUI, and run it through Docker. Literally download Deepseek. Yeah, check it out. Yeah, that's pretty cool. Awesome. Well, should we get into bookmarks, Brad? I know we're come up on an hour here. I figure we've got some stuff that we can chat about with the bookmark section. Yeah, you lead us. You lead us today. Okay, all right. My first one is actually, well, I will include the video we're just talking about. So it's host all your AI locally is the title of that video. So I will include that. Love that. But, and that was eight months ago. So he was pretty early on it, at least from what I know. But the video I linked in my bookmark section was actually not AI related at all. It's accounting. And it's from an individual named Logan Graf, who full disclosure. He runs a community for accountants that I'm a part of as of last month. But he's a solo firm owner, a CPA firm owner that specializes in taxes. And he's been really great about putting out like, I don't want to say he's like building in public, you know, because it's an accounting firm. But he's very transparent about like running his own. CPA firm, how he got started, you know, what his routine is and like how he gets clients. Like he's just kind of an open book, which everyone appreciates. And so he's been doing a series, I think since 2021, where he discloses how much he made as a, as a CPA firm owner. And, you know, the video quality is always really good. And he gives a full breakdown of like his costs, how much revenue, like, you know, based on what, like how much of it was tax revenue, how much of it was like community revenue. So he just put out two days ago his, how much money he made in 2024 as a fourth year CPA firm owner. And what was your reaction? Like how much was it if you remember and were you excited or? It's super cool. Yeah, I think he was around $2,000, $280,000. Really? Yeah, I want to. That's pretty good. Yeah. And it's, you know, and he works for himself. he, camera exactly what was hours It's not a side gig It's a full time thing No, no yeah Okay Time thing And let's see I'm I'm I had a video up right here It was 262,000 So last year He made 283 And remember last year He was saying He felt like he worked too much So it's probably I haven't watched The full video yet But I'm sure he's You know Probably saying Took less clients And wanted to work on Run in the community So that's probably Why that number is a little bit lower But you know Again as a solo firm owner I think he's got like one employee that just helps them with like part-time stuff. You know, that's all take home pay essentially. That's cool. I applaud anyone who's doing build in public and even like revenue numbers in public that takes a lot of guts. And I think as a viewer, I'm fascinated to see real like data that is usually not shared. And then I'm impressed at like people's grit to push through even through hard times. Sometimes it's you're sharing data that's not super sexy or exciting. Like maybe that number for him is disappointing. Maybe it's great. Who's to know? But not enough people share that. And I think more people should. Yeah. Especially because I think accounting is a bit more of a closed book than like tech is. And, you know, it's rare in my experience to see people be so forthcoming about like what it's like to run a firm. You know, especially because accounting, you're like, you're serving clients, right? So people go, I don't know if I want my clients and know how much money I make, you know, because they're going to say, oh, you're overcharging me. which is like such an antiquated way to think. So it's cool to see people like Logan that are, you know, brave enough to kind of put that out there. And, you know, face judgment, face ridicule, I'm sure. You know, he cares at all. But, you know, it does take some, you know, you got to just not worry about it and just know that people are appreciate. There's people out there that are appreciating what you're doing and that makes it all worth it. So, yeah, I'll link it in the show notes for my bookmark. And yeah, I'm excited to kind of finish. that video. I haven't finished it yet. Nice. That's cool. Well, mine is not as exciting. One thing I found on Twitter recently was there's a guy, Ryan Vogel, I've seen his name somewhere. He must be involved in some JavaScript community, like maybe a package maintainer of sorts. I apologize, Ryan, if you're listening and you make some cool stuff that I use, but he posted a video on Twitter the other day that was highlighting a Chrome extension. And the Chrome extension allows you to essentially craft your resume with AI by looking at the job description on the web page. So you can imagine if you open up your browse and you're looking at a LinkedIn job posting, there's a right panel that hovers and allows you to select the job title. So you do this with your cursor, which is pretty cool. You click on job title and his extension, and then you click on where that job title exists on the page. That will then extract out the data. Then you choose job description. And then you choose company name. So it probably takes you, and his walk through like 15 or 20 seconds to identify key pieces of information that, you know, what is the job title because those are the skills, job description also outlines skills and responsibilities and then company names you can tailor it. And then once you have all those pieces selected as Chrome extension, you click like process. And this button will take a look at your resume, modify it to kind of fit the skills that they're looking for, and generate an updated resume and an updated cover letter. And so you could apply to a job using AI, I don't know, 30 or 45 seconds. Yeah, it's probably not perfect. And I think AI models are getting there where they can manipulate your resume content to make it sound better in a certain angle, which for software engineering, you can be, you know, a mobile app developer, which kind of is my role. You can be a web app developer. You can be like a low-level systems programmer like working on gas stations or cars or all these different things. And from people who aren't familiar with the software engineering space, I can all feel like one. but when you get down there just like any specialty, there's so much difference in nuance and skills required in each of these fields. So if you are applying to a job, I don't know if he's actually posted this publicly, but if he does release it, it seems like it would be a good way to get a head start on crafting your resume correctly because the whole job market's pretty competitive right now in any edge you can get. And one, applying to more places because it's speeded up and two, having your resume crafted to kind of solve their story of the candidate that they're looking for, I could see it be a big win. And it kind of reminds me of LinkedIn's Easy Apply where they give you one button and it submits your resume. This is like crafting a resume and helping you submit. It doesn't go that far. But it seems quite valuable to have something like this as long as the AI model was good enough. So if your job searching and Ryan puts it out there, I'll definitely follow up on the next pod. But a cool video that I'll link, nonetheless, of a cool interaction with a Chrome extension. I've also thought about doing something with SME where you can log in to like Chase, You can have my Chrome extension. You could click on an expense and auto import it into my system. And seeing his Chrome extension really gave me an idea of this could be possible, but I don't think people love Chrome extensions, be able to read data on your screen, like access the web pages. So I was like, I don't really know. TBD for that. But really cool video nonetheless and a nice use case of AI and trying to stand out in the job market. Yeah, I'm not a fan of that, to be honest with you. Cool of that he built it. But yeah, I think people, like the AI tools, like as people, as someone who's hired in this era, yeah, you just know people are just like hitting the button and saying, okay, they call me back, you know. But I get there's, I'm not dissing the product. I'm sure it's cool. And, you know, for people that are, for people that are applying with actual intent, I'm sure it helps them out. But people that are just like. Don't spam apply. Yeah, exactly. Yeah, exactly. So that's cool. Awesome. All right. Well, another good one in the books, Brad. Yeah, we will upload this and catch you next time. Sounds good. All right. See you. See you later. Bye. Bye. Apple Podcasts or Transistor FM. To connect with Brad, you can go to Bradley Bernard.com or at Bradley Bernard on X. And to connect with me, you can go to Ben at Bernard.com or at the real tech CPA on X. You can also find us both on LinkedIn. Thanks. See you next time.