Inside the AI Engineer Role: Tools, Skills, and Career Path - Ruslan Shchuchkin

Hi everyone, welcome to our event. This event is brought to you by Data Dox Club which is a community of people who love data. We have weekly events. Today is one of such events. If you want to find out more about the events we have, there is a link in the description. Um you can check it out and see everything we planned. Then there is also a subscribe button. If you see subscribe, you should click because this way you will subscribe to our YouTube channel and get uh notifications about all future streams we have. Then last but not least, we have an amazing Slack community where you can hang out with other data enthusiasts. The link is also in the description. During today's interview, you can ask any question you want. There is a pinned link in the live chat. Click on that link, ask your questions, and we will be covering these questions during the interview. And I'm going to stop sharing my screen. And I already see some of your answers. So um Luca uh says that they want to they worked they work as a data scientist and want to become AI engineer. That's really cool. So um right now I'm opening the questions that we prepared for Rouslan and we're ready to start. — Hi. — Hello. — Are you ready? — Yes. — Okay. So today we have Rlan. Rlan is a genai engineer at finance guru where he builds production systems around large language models and generative AI. Before moving into AI engineering, Rlan worked as a data scientist at smart steel technology technologies and oils group. And um this is not the first time we have Rlan on this podcast and I know Rlan pretty well. We've known each other for I don't know how many years. Quite a few. We worked together at two weeks and at some point um Rouslan was a guest where we talked about biohacking, right? It was quite an unusual but also interesting episode. Uh and today we welcome Rouslan to uh join us as a guest again and today we're going to talk about AI engineering. So welcome. — Thanks for having me. It's always a pleasure to chat with you Alexi. Um so this week we'll talk about AI engineering and um before we start talking about like all the main questions I want to ask you to tell us more about your background can you tell us about your career journey so far? — Yeah sure. So my career journey actually doesn't start with data topics. I finished business administration program for my bachelors and then I worked in like business roles like account management or customer success manager for a few years but I realized at some point that it was a bit too boring for me and I wanted to learn how to build stuff myself and that's when I decided to try to transition and break into this data world and then I joined the master's program that dealt with it and I did some machine learning projects in the meantime and that's how I eventually got my first position as a data science train trainee or intern at Olex. — Okay. Um then what happened then? Then I worked at Olex as a full-time uh as a data scientist and I did some cool projects there and that's where we've met each other and um since then well funny part is that point Alex was also laying off some people and I got affected as well and then I was looking for a new job and um I landed a job at the smart steel technology where I worked as a the title was data scientist but basically I was machine learning engineer because I developed machine learning models for uh production of steel which was very interesting research field and I learned a lot there and back then ch uh came out and I was playing a lot with it and I got really passionate about this genai thing and I started doing my projects back then and um since then I did quite a few of side projects. — You mean personal projects, right? — Yeah, exactly. like I wrote a telegram bot that would help my mom practice her English and some other stuff. And um after I worked at Smart Steel, I decided that I want to join the AI engineering roles where I can actually deep dive deeper into LLMs, how you make applications for them and um how the users can interact with those kind of products because to me it's very interesting as a new domain. So that is my journey into AI engineering so far. Okay. So, you started your career before data as account manager, customer success manager. I'm just curious like was it in any way helpful in your data career now as an AI engineer or before as a data scientist? — Yeah, 100%. It's uh interesting how non data non-technical skills actually steer your career as well. So for me being an account management in account management I learned how to communicate clearly to set expectations build trust with people and that is very important whenever wherever you work with other people. So it helped me at lex smart steel and in finance guru now to talk to different stakeholders understand their needs their blockers how can I help them how to build trust with different stakeholders. So those are all transferable skills that are just important in life not only in your career. So I'm thankful for those years — and do you have any recommendations for people who want to improve in this uh communication setting expectations all these things like one thing we can work as account managers but not everyone has the luxury of spending a few years doing that like if I work now as a AI engineer and want to get promoted to senior AI engineer what kind of things I can learn to become better at these things — so it's actually a lot of like practicing a few things that I think are important. So the first thing is try to be honest and have a have integrity in your communication. So say what you think, ask people for their honest opinion and always be real with others. And this is how you're building trust with them. And then sometimes thing don't things don't go as you planned. So you need to also manage expectations of other people. Sometimes it means you need to say no. bring some not so pleasant news forward. But again, if you're being honest and you're being real, uh that builds trust and then you might lose one account, but you might win it in the future back because you keep the human relations and that applies to job search to many other areas. Just be real, be yourself, be clear with your goals and how you can help another person achieve theirs. — Yeah. say what you think um didn't really work out for me at school like teachers hated me um I hated them back but like I think eventually for my adult life it turned out to be quite a useful skill even though like it sometimes pisses people off probably there is also like a fine line how exactly you say what you think right like how direct you are it's also um I guess it's another topic u but um here I mostly wanted to talk to you about um your career uh your recent um stages of your career like becoming a engineer. So you said you worked as a data scientist and then you worked like your role was more like an ML engineer even though you were still called data scientist and then chapd came you were fascinated and you started building small projects like first project you mentioned was uh you wanted your mom to help learn English right? — Yeah. — What else did you build?

uh I also was fascinated with the chat GPT itself and uh how you can communicate with LLM and I realized that for example the interface itself is a bit too linear for me and I don't like it and I built another system where you can branch out of the chats and I called it branch GPT you can actually check it out branch/gptt. com um where — GPT ch implemented this feature They implemented it a bit differently than I did. And I like my implementation more. So in my version, you can select specific text that you want to branch into, not just the whole message cuz one message can have so many points that you can want to ask about. So for example, that was another actually a very interesting project because this was my first web app that I built. I needed to set up the back end. And AI engineering for me is more than just working with LMS but also understanding like and context management but also building end to end systems. Uh so that was another thing that I did — and I think I remember having this conversation with you and you told me we had a lunch uh about this project branch GPT and this was a few years ago right already. So when we didn't have these tools like clot code where that can write all the back ends for you don't even need to think like you give a prompt you go have some tea you come back you have a working up like back then we didn't have this luxury right so you implemented most of it yourself right — no it there was a moment when um when give me a second what's it called bolt new came out — which is also this autonomous agent that builds it for you and it did the very like the early version was built by that but of course I improved it later with like aentic uh engineering and wipe coding. — Yeah. So now it's like reliable and works proper but yeah back then I already used some tools because I think it's also part of my job is to be on top of all the latest tools as engineer and like that's why I like playing with them and trying to make something useful with them as well. So yeah, I already I'm going to ask you about this but later uh how you stay on top of that but first I wanted to understand like how exactly you transitioned to this role. So you built a few projects like telegram b um and then at some point you understood you realized that you actually want to uh do this full-time right so how exactly what did you need to do to make this transition and uh when was it? It was more than a year ago, right? — Yeah, exactly. A year and a half. — Back So back then like this role AI engineer wasn't even really there like it was only we would only start seeing these kind of positions, right?

right? — Yes. — Interesting. So can you tell us more about this like how exactly the transition looked for you? — Yeah. So just for the context, I was working for smart seal technologies and we did lots of like traditional ML but also very heavy ML. needed to do lots of feature engineering and like deploying these models and we had lots of challenges back then but um I realized that AI engineering is just something that I' I'm more passionate about. I see it as an open field and that is not only already interesting but is accelerating as nothing that I've seen before. So I thought it would be great to jump into that ship because it's just very new. So you can build expertise, you can build it fast because nobody knows what to do. And I was also pulled to the fact that nobody knows what to do with it and I like discovering like these new fields and uh coming up with solutions. So in order to do that I started uh I thought that I would actually do some more side projects. So I have specific journey experience but I was reached out by my current manager at Finance Guru who said that like we've been in touch um uh with him back when I was hired at Smart Studio Technology. I also got an offer from Finance Guru and we had such a strong connection on like how we approach building projects. We're both very pragmatic, very yeah like very agile I would say into building stuff. So because our philosophy about ML and data science and then Gen AI is very similar, we thought it would be great to work in the future. So when I realized that I want to explore Genai, there was the first company that I reached out to and then after some few discussions, it was obvious that they also want to start a team in Genai and I was basically the first choice and I was very like humbled and very felt very honored but that's was my way in. So it basically started one and a half years before I was looking for that job just because I established relationship and uh — how did the interview look like? Uh I mean my future manager asked me what are the projects that I did at the time and I answered and he asked me some very basic technical questions but he said that it's it was such an open field you don't know much about the requirements yet like he asked me about vector databases and some I mean agentic stuff was not even back then relevant people talked about racks so I answered some of those questions because I'm I was on top of these things, I was following everything. And he said, "Yeah, that's enough. You're more hiring for a like a certain characteristics, I guess, and some energy and drive and passion and — for cultural fit. — Yeah, exactly. And some track record that the person can deliver and actually build things — than specific requirements because I think that is what's going to matter much more in the future than just specific skills. — Mhm. So um in a way I don't know if it I can call it luck but more like um you already had established connections and this is what you capitalized on right — so I wouldn't necessarily call it luck cuz you actually did build this connection so it just didn't happen accidentally by chance right but because of this connection you managed to get it how would you suggest people to also build these connections with people even though maybe now they will not get the job but eventually in one or two years um they will have these strong connections that it will be easier for them to get hired. Well, I think that uh I would recommend everyone to just do their own AI side hustles and uh just build projects that are fun for you, especially highlight fun because if you build something that's not fun, you're going to burn out quickly. And if you're doing something that's fun, you will do it for a long time and eventually you'll get better. And either you can maybe monetize this project or at least you can put it in your resume. And I would also suggest to just apply and uh try to be yourself. I feel like many people when they apply there follow some patterns that you see on LinkedIn or um in other resume and you know you just put another generic caption to the resume and to me that doesn't resonate much so I would just ask be yourself try to show what your passions are — and just apply maybe you get a call with someone you talk to them you might not know but in two years they'll hire you as a j engineer — and that's what happened to That's really cool. — Okay, so — just like convert quantity into quality eventually, you know, try be yourself, have good intentions, and do something and — it is luck, but it only you're only lucky if you try stuff.

— There is a post I remember on the internet from the guy, his nickname is Swixs. He is also an advocate for learning in public. There is a post on how to maximize luck. So there is like what he the term he calls the luck surface right. So basically you want to maximize this l surface you want to establish as many connections as possible. You want to be as visible build as many projects as possible and you want to tell people about these projects and then um because your like surface is uh large. So these accidents that happen just by chance, you have more chances of this these things to happen, right? So I really like this post and you're living proof that this is actually like um something that works. — Yeah, 100%. I fully stand by it. I'll check out this post, too. — Mhm. I think I should find it and share it in the live chat. But um in the meantime, maybe can you tell us more about what you do at work? So you your title is EI engineer. What does it mean? — Oh, that's uh such a good and hard question. People have been asking me my whole career. What does it mean from data science to machine learning engineer? I would say that I identify a few um competencies that I have that maybe would stand me make me stand out a little bit from other software engineers or like data scientists. So I know how LMS work. That's one thing. And that means that it allows me to prompt them better, to build systems with them better. I have the specific ML skills that help me to fine-tune the model even though there interfaces for it. But I know that basically my tools my tool set is bigger and wider. So whenever there is a business need to solve certain problem because I have more tools to my knowledge and to my availability, I can solve this problem more efficiently or faster. Um and then I think as AI engineer I'm developing more into the full stack one uh soldier universal soldier that can make it all uh because nowadays the speed is what's most important I think uh all LLM enabled applications they're very new so many users have interacted with charge but they haven't interacted with voice so you need to build lots of stuff in order to validate what works. and what doesn't and for that you just need speed. So because — development speed you mean — yeah exactly and because thankfully thanks to cloud code you can develop much faster I also think that I want to be an advocate of code and similar tools to show people how to use them. So that is another part of my role that I see. So I would say product discovery being able to deliver fast and whenever we are at the point in the company where we need to productionize something then I have the skills like the full stack the system design overview to actually build it and execute it. Of course with the help of backend developers, devops, with help of designers, front- end people, um you need you know people to perform to use their expertise when it's necessary. But at least I have something to kickstart this. — How large is the company where you work? It's a like small size, right? Or midsize — around 200 people. — 200 people. Okay. I recently had an interview with uh Paul. Paul is also an AI engineer and he works at a small startup and um he u what he said that this is a very full like can I say very full stack like it's a full stack role you have to do everything and what you say kind of it's similar to what Paul is saying right so this is a full stack role um but what about the core engineering skills so you said like product discovery and all these things but like as an AI engineer Because what you say also a back end engineer can technically do use cloud code to deliver end to end application. They can also be an advocate of cloud code. They can show everyone um how to use it. They can do product discovery. But what is the difference? What sets apart? What sets you apart from back end engineers?

engineers? — Yeah. So to me the AI engineer is bec is someone who knows the latest things in AI engineering so to say because in back end you us you generally have certain patterns that are established and the person who knows and navigates around them comfortably is a backend engineer to me and while I can ask code how to do that some stuff that if there is just an article that has been published about there is another I can give you an example non-reasoning models uh follow instructions is better if you simply duplicate the instruction and because I know this and because I'm always staying on top of the field we actually implemented it and we see saw an improvement in some responses of the LMS we get so that is my expertise means that's the field that I'm constantly researching and that's why I can build the systems better even though nowadays with cloud code you know that can research very well for you everybody can do it so to me that is one thing so actually doing the core work of your engineer of enabling the LM giving the proper context um is something re realizing what has to be in that context uh doing the evaluations around it that is what I would consider core AI engineering but that to me comes a bit later once you have already established the product once you have verified that it is something that users need they see value in it and so on and until you have established that you need to focus on product discovery Okay. And it's just so happened that up until this point, this is what I've been doing. So once you have a proper use case, uh you would actually start optimizing again the prompt, you would optimize latency, you would optimize cost, maybe try smaller models and fine-tune stuff and do other tricks with context management. — Mhm. — So it all comes in stages and depending on what stage your project is, you need to exercise different skills. There was a recent post on Twitter, maybe you saw it from Andre Karpathy. Um, somebody saw that he's working on analyzing jobs. Um, like there's a database with jobs in the United States. So, what he did is at least this is what these tweets claim. I have not seen a post from him personally. So um he analyzed all these jobs and he ranked them how easily it is to replace this job with AI and the top ranking jobs are like uh software engineers and the the jobs that are least possible to replace are like uh rooftop um I don't know person who is like taking care of rooftops like plumbers like people who work with by hands and right now coming back to what you were saying about product discovery right so and I guess this is what everyone can launch cloud code like you don't even need to know how to program and say hey build me a website that is doing this and with the current models that we have they're pretty good at building things so probably like with oppus 4. 6 fix. It will work from the first attempt, right? And maybe from the second attempt after you. But what sets you apart from like anyone else is maybe this product discovery thing you mentioned plus the core AI engineering skills. U because AI cannot really do product discovery at least um from what I understand. Maybe you can uh confirm or uh not confirm this, but can you tell

us more about what product discovery exactly is and whether this is something we as humans should become better at to make sure we are not replaced by AI in the future or like how can we prove like future proof our skills? — Yeah. So I think that when you're building a product that is interacting with real users, you have an advantage as a human because you can understand better how real users behave, what they think, what they need. So for example, one of the PCs we built is like you can chat with your own data and um because I was playing with my own data, I could understand way better um what are the requirements and needs that I have. So that allowed me to identify what kind of structured outputs I need to do that the LLM might need to suggest me next best action as a part of the structured output and so on and that is something like coming up with features coming up with requests is something that I have while interacting with that model that I guess cloud code might struggle with I mean it's of course — it will come up with something but whether it's useful or not it's another question right? Yeah. And even if it would come up with it and if it would build it, you still need to show it to other humans to see how they react to it because I'm a biased uh person and we have real users. So for example, one thing we're doing with our design is that they are showing it to actually actual real users and we're showing them the proof of concept and we do this usability interview like how do you find it? we just observe them interacting with it and we come up with more features or realize what are the things we need to iron out so we can roll it out to more people and that is again part of being an AI engineer I guess wearing lots of hats um at this point is that is what we need to narrow down and really hit first building something that's useful and brings value — and designers and product managers are better at product discovery right so what if we fire all the engineers and just give cloud code to designers and PMS and then let them program what happens. I think it's a very valid point and I think that um well my take on it is that as one person you cannot be on top of absolutely everything. We still need to specialize because it is my passion to read about agentic engineering and I know lots of cool skills of clo code. I can actually share it with other people and that gives me an edge in developing better prompting techniques you know better agentic architecture and so on. Something that maybe I will do better than just a product manager with cloud code. So you still need to specialize you still need to know the domain in which you're working. My domain happens to be the field itself. But if you are a data analyst for example, the domain is absolutely crucial and no cloud code right now. Maybe in the future unless it talks to other stakeholders and collects the requirements and ask them what specific thing that they might need for something they don't even know themselves. It would be hard to do. So I still believe that the best combination is human plus AI. Uh not just human or not just AI. So I think we just need to give PMs access to more context so that their agents know yeah basically their context of work and they could work more productively. — Yeah thank you interesting point. Um and uh also these models they have knowledge cut off. So when I ask if I just start a project from scratch and ask it to implement me let's say I take a picture of something and I want structured output that describes what exactly on this picture right so what it will do it will probably use chat completions not responses API it will try to encode inject the schema JSON schema in the prompt instead of using structured output so it will do all the things we were doing two years ago not now right — and it might work it not work but it will certainly be quite behind of what the industry is currently doing and to me this is I feel like okay I'm actually useful I can now tell the agent hey like don't do it this way like check how to do this properly here and then reimplement and then come back — I think I will put push back on that actually I think that cloud code or cursor with a plan mode and web search actually fixes this so whenever I develop in a framework that I'm using or I use a new library. I first thing I do is I go into plan mode and I ask please research the documentation for my specific use case so that then it's up to date. I think there are even skills for cloud code and like plugins for cursor that allow you to check up the documentation itself dynamically. So you don't even need to worry about that. Um so yeah that is one thing and for the discovery of new stuff I also have some tips that I could share with you. — Yeah there's already a question um which is probably related to what you were um going with. Could you give us some

resources that you use to keep updated about new findings in the industry? But for me, it would be I would extend this question to also ask you how do you keep um yourself up to date cuz I open Twitter and sometimes it's so many new things and I'm like there's no way I can deal with all that. I mean recently uh last week it was just out research from Garpati in my entire timeline. I don't know what is so special about this but u at least there were not so much diversity but sometimes he open like this new tool somebody is building a side project and making like I don't know 10,000 per day with this and then somebody else is doing this and I'm like okay like it's a lot of information like so I have my personal way of dealing with this but what is yours — um I for me it's also Twitter uh that is my biggest source of inspiration. I also check hacker news sometimes and I follow some uh telegram channels. I mean some of them are Russian speaking but mostly 95% of all the stuff I find out is uh X and uh X for me is also about um just like the newest hottest thing right now. But when it comes to actual handsdown work and tips and choices, I usually go to meetups and I just talk to people and I ask them what do they use cuz I know that it might be battle tested. And for me, it's just more valuable to ask someone in person what do they use or reach out to them online uh and hear about their experience than just reading about it from a random person on Twitter that I've never met, never heard because I just have trust element to what the person I know says. Uh and yeah, you are right that there is a lot of noise, a lot of garbage information out there. A lot of — But even if it's not garbage, like it's just too much. — It is — like let's pretend it's not garbage. There is no fake information, which is not true. But like you open your feed and like all the links you see are actual valid things. It's still a lot. — Yes. So maybe one useful tip is uh I also buy subscription for X. Regardless of how controversial the whole platform is, I still find value in the subscription because it's the only LLM or chat that actually looks through the tweets, right? Because Twitter limited the chat uh limited access to that data. And because many people who are in the industry, they are talking on X. This is the LLM. This is the chat that knows the best. So, Grock is really good at finding the sort, you know, state-of-the-art models and whenever I need a certain approach like how do I do evaluations or aentic design in 2026, I just go on Grock and I asked there and then I know that it aggregates for my particular request the best practice. — Yeah. And uh John in the comments the same exactly the same thing that was uh his also experience with rock and my experience too like I also have a subscription and I really like XAI for its Twitter search or X search and Reddit search like when I usually say hey I have this question please restrict your search only to 2026 post from X and post from tra from Reddit and also sometimes hacker news because I'm interested in what people share. or not cuz right now we have so many articles that are generated like I don't know you open a medium article and you don't always know if a person wrote this themselves or not or there is a company blog post let's say I recently was preparing material for AI engineering questions and if you just Google questions for AI engineers you'll see a lot of posts on uh websites and you don't really know whether they just asked charg hey what is the list of top uh AI engineering questions and they just copy pasted it or they actually did some research there right so that's why I restricted to X and trade it and it works super well so it's totally worth it like this $8 I think this is how much we pay right it's totally worth it to me too — and I want to highlight once again the value of personal relationships you know if you are building stuff you're building in public and then you just meet people and they share their knowledge with you know I already learned about how to maximize luck stuff from you today that they have not heard about before. I mean, it was something intrinsic to me. I guarantee you that the more people you talk to, the more stuff you learn every day cuz that's maybe something they tried out and they found useful. So, to me, that's also a huge filter. — Mhm. Okay. How often do you go to meetups?

meetups? — Oh, right now because I moved to a new city very often. I actually organized my own meetup as well last week, which was pretty cool. and it was called AI Site Hustlers Club. So I wanted to specifically gather people who are doing some side projects and so we can learn from each other and I'm already learning a lot. So it's really fun. — I like I live by that belief that you need to establish connection. — Mhm. So how did you go about creating a meetup? — I mean you just post on meetup. com and people are pretty active on the website so they just sign up and u — so that was it. You buy subscription, you create a group, you set um an event. How did you go about finding a venue? — I had to walk around bars and restaurants to ask for stuff. I also called some co-working spaces, but they wanted to charge too much. So, I asked a friend who works in a restaurant if we could use their space, and then I talked to the owners and yeah, sure. And that's how we got the room. So, it was very cute and nice and very cozy, like non-corporate. You did you have uh like presenter with slides or it was just okay we get together and now let's everyone share opinion like how did you what is the format of the meetup? Yeah. So my idea is that again it's non-corporate, it's not networking, — no slides, no presentations. — So I want people in the future to like showcase their work. So again we can ask questions like what did you use here? How did you solve this? But at least for the first meetup it was just us getting around and I had a little speech in the beginning to say like what are we doing here? And in the future I want to have some presentations. Uh but we also want to do it online and offline like in the discord server online. So again, people share and learn together in public. And I feel like there should be a place for events like that in absolutely every city. Uh you know, so that you can just like to me building stuff like this is just fun and talking about it is fun, you know. I realized that it became my hobby just to do like side projects with AI and again like you learn about you learn lots of stuff. You talk to other people, you share same experience, — you know, and like make friendships. So it's just a really cool thing to do. I also saw that you are starting something like this and — yeah, — it's very nice. — I fully support that. Everybody should do it. — Well, in my case, this is something I want to do online, not offline. Um but yeah, I think it's similar. I really like the way you approached it. Uh because uh for me also like when it comes to offline events, the biggest struggle is getting a venue. And I really like your how to say lean approach, right? Just get a restaurant and then at the end the important thing is just get people together, right? And then start making connections and yeah, build something together or at least show what you're building. — Yeah, I think um that is something that I have like learned a lot is that perfectionism is a very bad thing. I don't want to swear but like I always need to push back against my perfectionism. So this meetup I organized with a friend of mine and uh we had so many plans. We wanted to make a website and we also wanted to make the presentation so we need the screen and eventually we just did the meetup and we just got people together and I think that the imperfect meetup that happened is way better than the perfect meetup that would never happen. So the same goes for AI projects. Like I try to scope my projects to be as small as possible and as imperfect as they are, some of them are complete and they're bringing some value and you know some people are using it and it's fun. So that is also like a very strong advice for me is just build something small, learn something with it and then move on to the next. — That's so cool. tell us more about the site projects you have right now. Okay. — Yeah. One of them is um a Chrome extension called Catch a Flat that refreshes Emoscout page for you. It's like a — Emosc for those who don't know. — It's a classified um like a place where you search for flats basically. — And in Germany only in Germany, right? — I think so. Yeah. Uh and but in Germany it's like really hard to find a place to rent. So this extension refreshes the page, the search page for you in the background. So you just like run the extension, it refreshes the page and whenever the new flat appears and it matches your search, you get instantly a notification on your Chrome. So that way you can actually catch a flat otherwise it would be gone because too many people apply. And I have like 150 users and I got three donations from it which is super nice. Like I didn't expect anything. And it's completely vibe coded. Like of course every now and then I asked CL to review if it's fine but I also open sourced it so you know it's like very imperfect but it works and just like two weeks ago I got a donation and the person said they found a flat with it. I'm like couldn't be better you know. — I just had a side thing for like a week — now you can buy yourself a coffee right? — Yeah exactly. I bought I wrote a bike club mate which is like this Berlin drink with coffee. — Mhm. — Yeah. And then I also am building now an app for my phone that I'm going to launch soon and it's called Phoneless.

And the idea is that you just spend less time on your phone. It sends you some funny notifications. Um so yeah, I'm like I'm working on many projects at the same time because I like to switch between them. — Mhm. Do you understand correctly that these projects they the motivation is coming from your use cases your struggles like something you want to cuz I guess you didn't think like okay what do I do let's build an extension for looking for a flat probably you had to find a flat yourself and you realized okay these are the limitations this is how people go about this and this is how I can automate it — yeah exactly yeah again like my goal with the projects is maximize fun you know and solving a problem that I struggle with is fun for me or if it's a problem that like my friends or you know my wife is struggling with like solving stuff for them and I can show them look it works like that gives me so much motivation and that's why you know before I thought like I should maximize productivity I should maximize you know like whatever the money I'm making with my side projects or like the stuff that I can put on my CV and when I maximize those I instantly burnt out and I was just stopped doing stuff. So, maximizing fun, even though fun is such like not the best word, but just, you know, passion like be driven about it is my way of doing it. And that's why I'm consistently doing lots of stuff for like a year. So, — yeah, just like do stuff that gives you passion. Switch between projects all the time, abandon them, whatever. Just do whatever. — I do this all the time. Thanks for the advice. Don't feel bad about it because you're still doing something all the time, right? So that's the main point. — Okay. — That's how I do it. That was my revelation once. — So I also do a lot of side projects. Um but none of them are making money. Um so and I was thinking like cuz you also open I don't know if you follow twe follow these people on Twitter. So some of the people I follow on Twitter they come from this indie hacker community. — You know this right? Um, a few years ago, was it 5 years ago, there was like a big threat of people who and we were sharing our Twitter accounts there and we would follow each other and since then I still follow quite a few people. — Mhm. — Um, and they're building side hustles. So, they're probably also maximizing fun, but because in their community, it's quite common to also share like what works, what doesn't, and they to share numbers, right? So they say, "Okay, like I build this website and this is how much I'm making on this website, right? " — Um, and I'm like thinking, okay, like I'm maximizing fun, but maybe I'm leaving some money on the table cuz this is not bringing money me like I spent some time, but I'm not getting um money back on that. — Yeah. like what's your you said from what you were talking about I had impression that you all also try to do something like this but tell us more about this experience. Yeah. So, um I think that if you want to launch a successful business, a project that actually does money, brings you some money, it requires a lot of different skills. You know, you need to come up with an idea, validate it. You need to design like a proper interface. You need to have a very clear product messaging. You need to actually build it to have a nice smooth fast front end or app. You need to have a back end. You also need to set up payments. you need to come up with marketing strategies, figure out the channels, you need to uh do the whole legal thing which is just another world. So there are like so many components to it and I believe that if you're doing a side project that teaches you one of those components then by project number 20 you would probably gain lots of skills that like my idea is that you know I I'm doing it as a side thing like these indie hackers they go all in and they mostly just go to like uh East Asian countries or wherever the cost of life is very cheap they have some savings and they just grind for like one year but that is not how I live you know I enjoy my work life balance. I love to spend time with my dear ones. So

I just do these fun projects and I deliberately think like with this project I want to learn app development or how to deploy stuff, you know, or I want to implement payments. And my idea is that first of all that makes me a better AI engineer, you know, because now I have a much better overview of how these systems work on a bigger scale. But secondly is again I'm doing something that might create me another source of income and as I do more projects and I follow fun then eventually I'll gain the skills to actually start something that would be like have a much higher potential for success. I really like your attitude and also what you mentioned is try to scope the projects to be as small as possible. So then you learn like one specific thing and if you approach this um you said starting a business you break it down into multiple skills and you think how do I learn each of these skills separately in isolation then by project number 20 you'll have like a good portfolio and probably these people indie hikers they already have successful projects so for them they already have a playbook so they already know how to take whatever works and replicate ated in the new domain which I don't have yet right at least like if I talk about SAS software service projects right um so it's um it's quite interesting what kind of possibilities we now have with all these idols — you know I think that again each of us has like a personality something authentic about you know and just as an idea like because I've known you for a while I know that your authentic personality is like bringing knowledge to people you know and that could be your main business So — or maybe it is already — it is already but I'm just saying that you know this is the stuff you feel natural doing and that is the stuff I think you'll have most success doing further you know like eventually it will bring you much more money so um that's why I want to emphasize like do something that feels good for you um — that you resonate with that makes you just stay a few more hours at your computer after work uh every single day you know after having a whole work day because it would not it would feel less of an effort — to actually do it. — Um yeah, that was my message. So, — okay. — Thank you. — Um a few questions. I see a few questions. Question from John. Do you have any experience with AI hackathons? — Uh I do. We did some inside the company and that was super cool and interesting and I actually really believe in hackathons because you have this dedicated time and everybody's super passionate and driven to do stuff and because you have time constraints you cannot do a perfect thing so you have to do something quick and dirty and I know so many companies and like people who started their startups or side projects that — originated in the hackathon. We're going to do some hackathons for my community as well. So if you can get your hands on any hackathon, just go for it. It's really fun. — Mhm. So uh for hackathons um and in general for projects, one thing I see in the community often is uh there are people for me I don't have this problem. For me, I always uh if I want to work on something, I always have like a pile of projects that I can choose from. Like I never have the lack of what to work on. But sometimes there are people in the community and I interact with them often that um do not necessarily have this backlog of ideas to pick up from. Right. Do you have any suggestion for them? So one suggestion could be to go to a hackathon and then they would just give you a project to work on or at least some idea they will push you to towards some direction. But if I want if I have a few hours per day and I want to spend on building something meaningful, can you suggest the source of uh inspiration for me to select a project? — Well, the approach that works for me is just do it, you know, do something stupid. do something you want to tell your friends about later and laugh about, — you know, — like just find this source inside of you that's like, oh, you know, I made this whatever outfit checker that just sends some images to D banana and builds me an outfit because I don't know what to wear or like, you know, — you have to have a problem, right? You don't know what to wear and then you want to solve it. — Yeah. Right. — But all of us have lots of problems, you know, so just pick one. And also many people think oh it has to be so big and huge and perfectionism kills so many projects before people even start. So just pick a problem ask code how to solve it let it completely vip code it but learn how to deploy it and show it to your friend and this is your first project. — Mhm. But how do I just pick one? Like what if I just I don't know I sit down I stare at the screen and um — maybe you're person to ask. No, I have an answer. You don't come up with an idea when you sit in front of the screen. The idea comes in the shower. It comes when you go on a walk. talk to someone. So, just have a phone nearby to write it down. — It never comes to me when I sit in front of the laptop. I run towards it to actually do something. You see? — Okay. — So that's why taking a break dissociating from your daily like from your work from your tasks it has to be it comes from lightness within you not from something like I'm now super deeply focused and I come up with the best idea in the world. No it should be simple and stupid like start simple that's my only suggestion. — Yeah for me the source of inspiration for projects to build is usually things I do. So I do something then I see that something doesn't work the way I want.

So there's a clear area for improvement. So recently I was running cloud code on my computer on my laptop but the problem is you close the laptop and it's not working right. So the solution is I was thinking what could I do? So I was thinking maybe I can run it on my old Android phone. Mhm. — So, I break the phone, so it didn't work out, but at least like I had fun destroying my phone, right? And then I thought, okay, I just rent out a computer and then SSH to this computer and then code is working there. But then I was thinking like, I don't want to necessarily open ports on this computer cuz somebody can hack in and rake my computer. — What do I do? So then u visual studio code has this automatic port forwarding you know like I don't know if you use it but like it detects that something is running on this port and it automatically forward forwards it to your computer — but what if I don't use visual studio code I mean I use it but what if it's not open right so I still want the ports to be forwarded and then I ask cloud code to implement a uh um a terminal app that is doing exactly that and it was just a problem that I had and there are so many problems I come across. This was just one of the examples that uh solves a problem that solve a problem I have right now. And I found out that there are so many problems I have right now. — Yeah. — Like the list is so long. Uh and it's like I don't even know where to start. — Yeah. I think that is the very unexpected outcome of doing projects is that you just come up with more project ideas all the time. Exactly. You know, so that's why you just need to start with one simple stupid problem that bothers you or you just want to try it out. It has to come from fun and from lightness for someone who's never done a project. But if you talk to someone who has been doing projects, all of them have like already 20 more ideas of what to build next. So — yeah, so you can talk to them, right? — Yeah. Exactly. Or just talk to them and ask, "Hey, what are you struggling with? " — They will tell you what to build. — Yeah. Actually, some of the students of my course are doing exactly that. they say I don't know what to pick up and I give them a project to work on. So for example, right now I want to create issues on my GitHub project through voice. So I just record a voice message to a Telegram bot. It determines which project is it about, what is exactly the issue and it just goes and create an issue on GitHub. — This is a super simple thing, right? — Um yet this is a problem I have and I don't have time to implement it because I'm working on something else. So the student just implement — Yeah. Great. — If Yeah. If you're looking for ideas, you can just let me know or maybe too. — Yeah. I will definitely drop you a long list of stuff you could do. — Yeah. — Um — Yeah. — Oh, no. I just wanted to say that I love working with Telegram. That's why my first project was in Telegram and I still have made so many bots like the bot that translates from German to English for me but in a natural way. And there is another cool way of doing it in Cloudflare that like you can basically deploy it for free and run it up to 100 requests per 100,000 requests per day. So Telegram building stuff highly. — Yeah, Telegram is amazing because um like it's so easy to create a bot. You just talk to botfather and you have a bot. — Yes. — Okay. Um I saw a question at the beginning when we talked about your career. You mentioned that you worked as a data scientist and you worked still your official title was data scientist but you were more doing like ML engineering but then while you were working on this you realized that you want to work more on genai stuff so you became an AI engineer and then somebody asked a question um do you think that

there is no much future in data science — I think there is definitely because again what do you define as data scientist depend depends on the company. Uh in some companies like Meta is data analyst who might do some uh forecasting models for example. Um but in other companies like Olex where we work data scientist is someone who develops machine learning models and deploys them and so on. So regardless of the definition I think that for all the definitions it's still valid. You still need to understand what you need to build. You still need to do some stakeholder analysis. understand what other people's people around you might need it. You need to properly have a clear product vision for that. So identify what would be necessary, what would be the MVP, how to properly plan out implementing it. So you could test and verify, run some AB tests, maybe have some baseline in the beginning. I still believe there is a space for those professions whatever they mean but uh definitely it's going to be accelerated by AI agents who know your context and who enable you to build stuff faster. — So it is just a reality nowadays especially it will be more relevant in half a year that you need to work with agents as your co-pilots core workers. So just jump in faster you could teach others maybe profit from it later who knows. — Um what about this perspective? So now with clot code any back end engineer any software engineer can do the work of a data scientist. So why data scientists are still needed if like if you just give I don't know max process friction to your back end engineer and they will train all the models or will they — uh they might but again I think that sometimes you need to spend time on the system to improve it. So yes they can build like some dashboards. whatever train a machine learning model. There was actually an article from entropic that claw fine tetuned itself and so on. So found the data qualif made sure that it's high quality did the finetuning deployed the model and so on. — And that's how we got the 4. 6 opus right. — Yeah probably. And uh you still need to spend time thinking about it. um you still need to identify what are better requirements for the system and improve it further. So if you just need to get something done fast and dirty using cold code is good but if you are especially working in a company and there's just a higher standard for it uh you need someone dedicated who thinks about it and improves it all the time and again verifies that there are improvements. So I still believe that we will need all these people — right I'm one of them but um you just yeah the quality will be higher — from what I see data scientists have multiple superpowers that engineers don't necessarily have one is this business acumen that engineers usually engineers are more focused on uh building things and less focused on this uh how did we call it like product discovery Okay, which is fine, right? This is eventually this is something that PMs do. designers maybe analysts are more involved, right? And data scientists in my experience, they also have more this business exposure than engineers, right? So this is one of the superpowers. So data scientists are better at translating the requirements from business to uh machine learning terms and same with AI, I guess, these days, — right? And another thing is what data scientists are really good at is evaluations like for engineers this is more like a foreign term or more foreign term than for data scientists but as data scientists we have been doing this for a very long time right so we know what it means to evaluate a machine learning model and for AI it's basically the same thing more or less right the tools are different but the approach is the same — yeah 100% — yeah um do you have time for a few more questions — yeah go ahead — okay — sure — um How about getting into the industry without a degree? — Um, hard question. I think you need to maximize your luck field again. Uh, talk to people. I mean, I know that some people join like a software engineers without ever getting a degree. I just met a guy recently who did that without getting a degree. So, he had some side projects that he did and that was good enough for the company to hire him. So, just do side projects. I don't see any other way to be honest. I think without like especially if you're only starting out uh getting an internship is also a very — valid way but I haven't been looking to be honest at how many internship positions there are posted if it's still easy uh to get it. — So I would just say do your side projects. They will definitely be useful for you in your life. at least you have some stories to tell and at best they could also lend you the job. So — and I remember when I interviewed

people I never really cared about what they have in the education part I never actually even looked at it right so because like at the beginning uh you have like some sort of summary then work experience and I don't even scroll to the education because like I couldn't care less. — Yeah. — Right. Because for me the important part is like whether they can do the job or not. And this is what I'm testing during the interview. But I don't really look at the education. — Yeah, I totally agree. — Whether it's MIT or the university I graduated with in Russia, nobody knows about it. It doesn't really make any makes any difference. Well, maybe in MIT people are not maybe like people are definitely smarter there. But then my goal on the interview is to determine whether this person will be able to do the job or not. And while there's probably some correlation between having a degree, uh I don't look at this, right? — Yeah. Vid I would also wouldn't necessarily look at it. — Mhm. — myself. — Um when working on personal projects, one of the issues I face is I run off tokens very quickly. What do you do? — Um use plan mode. uh of actually I have a really cool tip. Use templates. Many people start projects and they just start out with like build me a whole front end from scratch and then obviously in order to generate lots of files like you're going to burn lots of tokens. Why don't you just look up at the template from Nex. js or something like this that would save you a lot of tokens. So that's one. Second, um, heaven when I tried to think clearly what it is that you're trying to build. For example, when working with front end in the past, I just burn through lots of tokens because I didn't really have an idea of what I want to build. That's why I'm trying to learn Figma right now because it's apparently easier and faster to iterate there. Just like have a sketch, have a think about it, spend some two, three minutes, talk to quote code superpowers has a skill called brainstorm that allows you to brainstorm very well with AI to really narrow down what it is that you want to build and then give it to plan mode so that it it discovers what to build and how to build it fast. And um yeah, I think that using already existing code just reusing it with templates is good and having a clarity. — What lovable is doing, right? This is what bolt new is doing like all these uh bootstrap I call them project bootstrappers. This is what they do. — Yeah. And maybe another little tip that I'm using a lot recently is that to get this clarity of what it is that I want to build is actually really hard especially if I need to put it in words as a prompt. So what I do is like I just start the voice mode and I just completely dump my brain into it for like five minutes. I try to look at it from all the perspectives. Then I ask please generate the summary and like structure my thoughts and then I give this as a prompt. Um so I don't have to spend time thinking and properly phrasing stuff. I actually you know let AI do this job. — What's your workflow? Do you just uh create a new folder, start cloud session there and do your brain dump or you first use know web interface to actually understand what you want to build and then you start a cloud code session. — Uh right now I just use code session right away and sometimes I would just like copy like I did the Figma plugin a few weeks ago and I just also started from a template and then I used the plan mode and just dumped my brain into it. Um but uh in the future especially when building interfaces because that's like currently what I'm a bit passionate about. Uh I want to learn Figma. So I would actually start at least doing some very easy highle mockups. You can also just they're useful because you can pass them to cloud code and to implement based on that — and yeah uh Figma released MCP recently. Um but also you could just keep screenshots that would work as fine. So yeah, having — I use the $20 one. I mean, at work I have the $100 one because I use multiple things at the same time, but at work with 20 bucks, I think it's also I like that it's limiting me in how much time I can spend doing it. So I try to actually have like a deep focus like think clearly when I'm using it and also when I'm out of credits, I just, you know, take a break and go play guitar, like talk to someone. Mhm. — So, — do you use Codex too? — Uh, no. I cancelled Chpt a few months ago, but I think I will take it again. Like, I'll buy it again because I had — $20 you can get way more in Codex and include. — Yeah, like I'm still using cursor and I tried all the models because I feel like to be on top of things, I need to try everything. But — Mhm. — like it's I still struggle to what exactly I would recommend but probably cursor uh because you have you can try everything and it's a bit more visual for non-coders and for coders just go with quot code. You would get a lot from just $20 a month. — Yeah. Also I would recommend GitHub copilot. It only costs $10 and it give you gives you way more than $10. Right. — Okay. — So — Okay. Last question. You have time for one more. — Yeah, sure. — Okay. — So, how do you find uh um like now that we have this AI, do you think you learn more building with AI versus previously?

Cuz maybe we can go deeper now with things and we just learn I don't know prompt engineer rather than something else. Um do you think AI is helping us learn or actually preventing us from learning? I think that the natural way of things is that uh people do not learn. They just wipe code and let it be. So I had to realize that and then act against it. So now I'm trying to actually learn from what I code. So uh if it's a production system, I need to make sure I understand every single line of code that's being written. I mean for my personal projects, sometimes not, but for the work ones I do. And I think that is again a very powerful technology to let you grow your skill set. Um so yeah just to TLDDR uh usually people don't know exactly what's happening when they vibe code and I would recommend take your time ask exactly what is happening on every line also like rename things so that you have a clear understanding when you look at the code and um yeah learn more because this is the best the fastest way to learn right now with AI than ever — and how do you decide whether you want to learn more about a project or Because for example, I have a project. You know what Jackel is? — No. — Jack is um it's a static website generator written in Ruby. It's used in GitHub pages in almost all GitHub pages. So it's written in Ruby and my site is very slow now. It uses JL. So I decided to rewrite it in Rust. I don't know Rust. So I have to fully trust the LLM that it's doing the job right. cuz I don't even have the desire to understand what it writes because it's Rust. The only thing I learn here is what exactly you need to do in order to build a Rust project. So I'm learning a little bit but not really like I cannot say I learn Rust even though I have a project written in Rust that seems to work. — Yeah, I know what you mean. I mean it's the same for me with the TypeScript and um yeah I do some stuff in Cotlin too. So I have a very comprehensive cloud. md for example that I refer in cursor as well. So I at least need to make sure I understand on a high level what's happening inside the code and to me that is enough. So I know roughly how the app structure is what are the files that I have there and how do they define you know how the behavior is and so on. So um I think you need to dig deeper only if you have to you know and if you want running something that is like mission critical then make sure you understand every line if it's just a project on the side again — if you don't need to go deeper and it works that's fine with me. — Mhm. Yeah. Okay. I share this attitude to be honest like sometimes I really want to understand what is happening but with this project I think like I actually wanted to learn Rust for I've been wanting it like for last three four years so now I can finally create Rust projects but I still have no idea what's happening. — Yeah I mean like us knowing one programming language is an advantage because I'm learning TypeScript now and I just do it with AI. So claw generated me like a whole how do you call it like a study plan and for all of the stuff that it describes it shows me how it is related to Python you know concepts so for me that is like a much more graspable I don't think there is a TypeScript book for Python near like Pythonistas so that really helps me learn it faster and that's what I would recommend you know if you already know one language learn other stuff in relation to Okay, I think I've kept you long enough. Was really amazing to ask you all these questions. Thanks everyone for asking these questions. Amazing discussion. Thanks a lot Rlan for joining us today second time. Um I would be really interested in catching up again like maybe in a year. Um cuz you've been in AI engineering since it kind of started it kind of formed. So it's really interesting to see your perspective on things. So, thanks a lot for doing this today. Oh, — it's my pleasure, Alex. Yeah. Thank you. — Okay. Well, I guess that's it for today. So, everyone have a great week. Um, thanks for and see you around and enjoy the new city and uh