The Real Difference Between Data, ML, and AI Engineers

Automatic transcript of the recording (English); speaker labels are automatic. Timestamps open YouTube at that moment.

0:00Sacha: Welcome back to AI with Sach. Today, I have the pleasure of hosting Og. Og, you do so much, you've been in the tech space for a while now. I think the better way to introduce yourself is just to ask you, who is Og?

0:14Guest: So, who is Og? Well, originally, I'm actually an engineer. I studied computer science and became an engineer a long time ago, 25 years ago. And I've been doing this for around 20 plus years. I got into the data space in the last 10. And I am somebody that loves to tinker with technology. So, it's not just about data. It's not just about software. I also like playing around with hardware. And yeah, I consider myself a tech enthusiast, but also a data professional.

0:46Sacha: And by the way, thank you so much for joining me today. So, Og and I work together from a distance. We've been conversing about software engineering versus data engineering, etc. My first question, because you can help, and I think a lot of people would benefit from this conversation. My first question is, what are the core responsibilities of data engineers versus data scientists versus software engineers, ML engineers, AI engineers? We have so much nowadays. And I would love if you could clear up a little bit about the differences between each. I

1:23Guest: do have to say this, like, a lot of these perceptions of what these roles are, are very dependent on where you work, what that culture entails, and what the expectations are, right? Because, and even the level of maturity within a company is very different, right? Like, you cannot compare, like a CEO from a startup versus a CEO from like, Apple, right? Like, it's a very different thing. Same thing with like a data engineer at Apple. be very different as a data engineer in a niche boutique consulting firm, right? Like it's a very big gap there, but there are certain expectations or I think ways of looking at those roles that has become like a common understanding, right? So to clear up a little bit of how I see these things, data is going from usually, data goes from a transactional way of looking at it, which we call OLTP or online transactional processing versus OLAP, which is online analytical processing, right?

2:27Like it's very different. One's a real world representation of what is happening versus OLAP is analyzing what happened and trying to get insights. So those are kind of like the two sides of data. Let's start with data engineers, because I think that one's kind of like the base. Data engineers are kind of in the middle. They're the ones that are translating, right? They're translating data for what happened in the real world and where it got saved. It could be structured and unstructured and then making, moving that into a form, right? That is going to be usable for the other roles that are involved in data, right? So for example, believe it or not, a software engineer is on the transactional side because they're interpreting ways of ingesting data from real world processes into a structured form or unstructured form just gathering it right and then once that happens the data engineers go and grab that information transform it and land it in in a way that is usable for somebody else data scientists ml and ai folks are the ones that consume that information in all of their process and all of the that journey that goes from olgp to olap they're consuming not just the data but the metadata of the data to kind of infer and understand what's going on and try to predict and and do other things right so that that's kind of like the base way simplified version of of that

3:55Sacha: where do you see that the skill set because there is a lot of skills involved in all of the different jobs or the different roles right where do you see those skills and those skill set overlap and where do they kind of complement each other or completely different so

4:13Guest: at the end of the day all of them are handling data right so when it comes to tools and skill sets i would kind of like separate a little bit of how those overlap with with uh um the five roles so let's let's define this data engineer data scientist software engineer ml and ai engineer right like that's going to be the base of of our conversation here um the overlap here is they're all using tools the tools that are being utilized might be a little bit different so for example a data engineer might use like something in spark or understand how uh like kafka would work which is uh ingesting real-time data or cloud data warehouses right that those are the kind of like the tool sets that you have it could be enterprise tools it could be sas products it could be with different programming languages some of those tools can be utilized by a data scientist let's say databricks as an example it's a really good one the data engineer could use pi spark to move data from one point to another and transform it while a data scientist uses the same tool with different techniques to start analyzing and creating insights right so it's a it's it's kind of a little bit of a i'm trying to find like an analogy so carpenter and a builder they both use a hammer but they're doing a chair the other one's building a house i

5:36Sacha: do love that analogy actually when we look at a pipeline sometimes people get confused who owns what you know when we go from ingestion and you've touched upon that but i want to go a little bit deeper so which role kind of owns what in a typical ai pipeline from ingestion all the way to production okay

5:56Guest: this is more of a depends on how you work a situation or how your team's structured internally and how you split those responsibilities right like i've seen places where the data engineer is doing ai ml and helping out with certain pieces of that first and and kind of like being the devops person for a data scientist or a ml engineer versus when it's with a software developer it's understanding their data model how they're saving data how often and understanding all the data to be able to generate like schemas and whatnot and that's like a very different uh you know um dynamic so software engineer would be the ingest well not the ingestion piece but like landing it somewhere that is going to be think of like the back end of a software system right that's their responsibility data engineers are responsible for the framework where they're running their data pipelines so let's say databricks and workflows and all of that or let's say a data warehouse and etl done with pure sql would be another like they're they're responsible for for that pipeline data scientists are more on the semantic layer right like so once the data is landed in a specific form you're you're creating another like layer on top where you're transforming or changing data or doing some calculations on top of that that's what those folks are in charge of ml engineers are going to be specialized in training machine learning models and i think ml engineers are the ones that do the most for themselves because they're doing ml ops they're doing like a little bit of they're kind of like data engineers but on the ml space i i kind of see that that same same thing with ai engineers right they they have like an end-to-end view of the whole pipeline right and they have to understand software development they have to understand data and ml and in data science as well they feel kind of like the bridge to do uh end-to-end because they're building ai solutions right it's a little bit different like doing a rag model or uh figuring out how they're going to set up a vector database where that's going to go right those type of things right so it's very difficult to kind of delineate where where each person goes because again this is new like i was right on i was doing data right at the dawn before even data engineer was a thing right so it was called like a dba or it was called something else where like data engineer just changed so much during the last seven years where i think it's been like seven years since it became going to turn basically people just spin up and say whatever they want oh yeah that person's a data engineer and all they're doing is i don't know like drag and dropping some boxes and and that's it and etl well you know it's a little bit more complex than that right

8:51Sacha: and i love that you said you know all of this is a bit new and specifically for me ml engineers and ai engineers are picking up steam and they're becoming a little bit more popular nowadays how would you differentiate between an ml engineer and an ai engineer okay

9:08Guest: going back to the antenna intelligence systems yeah ai engineers that's what they do right they're creating the like how are you going to adjust the prompt what type of uh security you're going to put behind it what what different types of i forgot what there's like not techniques but more more like i'm going to say models that you could put in place or architectures for determining if there's a prompt injection or uh if if the ai returns some uh hallucination are you going to return that back in your system or do you have like in the agentic uh architecture behind you and how you're going to expose that back to the user like that's a little bit of a different for versus like an ml engineer and again it depends where you work uh you're training you're tuning you're optimizing a specific uh ml that's going to help you a specific model that's going to help you do like fraud detection or just more like specific than a open-ended business process that's sending you information where you use gen ai or some other form of ai to kind of return a response so they're a little bit different right i

10:17Sacha: agree and also something that's maybe interesting for folks listening if they're pondering which role might be a better fit for them right which of these roles do you see as being more r&d heavy versus a little bit more operational like production based or production focused so

10:35Guest: let's look at the wording on each one and we could like infer really quick so you have the data engineer you have the software engineer and the ml engineer those three are going to be very operational heavy right we're we're the guys there like fixing the pipelines like like plumbers basically and trying to determine what's the fastest way to get things done right like those are the the three data scientists and maybe ml engineers and ai engineers could do a little bit more r&d heavy due to the nature of the of how it is right now right but like everything everything starts this way it's very r&d heavy to begin with once it becomes mature enough as a role and becomes mature enough in processes and becomes mature enough in best practices and regulation and all these other things it becomes more operationalized as it evolves so for now the engineers are the ones that are creating the pipelines the ones that are doing like the base work and the people that use that data on top are the ones that are actually doing the research and development and

11:38Sacha: now that you you're working in consulting you see a lot of projects and you see a lot of teams being formed right to complete those project what does a complete or successful kind of ai team or data team looks like in practice so

11:54Guest: that one's a very hard question because i'm going to give you the consultant answer depends so it depends on the project man one of the things i've seen that works the best is when you have first of all people that are interested in working in investigating a little bit more right like that's what i've seen like yes you can be a very good data engineer that knows how to do a pretty good pipeline but if you're not asking the questions to not just the stakeholders but to your team and try to understand how you're going to solve that problem even if you keep doing the cookie cutter thing it's not going to work for the particular project but a good composition or what i've seen in different projects in the last like 10 years years of doing this, having a solid data engineering team that understands software engineering principles helps a lot. When you have a data scientist that knows how to explain what they need in the very, let's just say layman terms so that you could get requirements gathering and then push that back to the data engineers, that's very important.

12:58Having a, and it depends if you're doing, let's take like a machine learning operations because that's what I've seen the last year's AI engineering. I haven't been in a project where they were doing like a, well, only like hackathon or POCs for AI engineering. So I'm going to focus on ML for now. Having a good DevOps person that understands data is important. I know we haven't covered that role here, but I do think in the next five years or three years, we're going to start seeing more data ops, right? Because we're, we're, we're, we're in the AI era now we're in the, we pass from the information era to the AI era. And what does that mean? Information is important. It's, it's basically what's going to drive AI to do what it does. So having a good infrastructure that is compliant, that is secure, that is not going, that is easy to use and helps people and enables the team that is going to be very important. So I would, I would throw in data ops as another role in, in here.

13:58The teams I've worked at that have been successful have been teams where the end user or the machine learning or data science person know what they need and know how to explain it. You have data engineers that have good practices to speed up and maintain a system of quality and a framework of quality is important. And then having a robust data ops person or team behind you is also important. Right? So I think when it comes to like tactical, right, like the tactical piece, I would look at that the strategic is very different.

14:32Sacha: And I love what you're bringing up. And what I'm hearing is that more and more, what's going to be important is always is how can my engineers become more fluid, right? So how can a data engineer can also become aware and asking the right questions and then the DevOps person be able to understand data. And this is a very, very interesting topic and something that, that I've noticed as well is, as we move towards this AI age, as you mentioned, it seems to me like the people that are going to be successful and the teams that are going to be successful in companies that are going to be successful are those that can be, that can adapt very quickly to moving technology. So I want to get your take on that and hear a bit about how we're moving from very specific and detailed roles, right in the past to more fluid, maybe teams and more fluid roles in a company.

15:22Guest: Yeah. I'm going to tell you the, like one of my answers, and then I'm going to answer with a little bit of a speculation, if that's all right. One of the things I noticed with, with this, and yes, I do agree. The future holds us being more fluid. I've noticed people that know how to adapt themselves and know how to ask the right question are the ones that are successful. And with the way AI is going and how I, I have a bit of a hunch, if you want to call it that companies are going to start creating their own LLMs. Just how would they have their own data warehouse? It's the same purpose. If you tack on top AI, it just enables people to ask questions in natural form, to be able to get the insights and, and kind of throw ideas back and forth with, with some, something that can reason and can understand what you're trying to ask and go back and forth and create something. So I do think being fluid is going to be very important. Somebody that's a, is not afraid to try things out. I think that's, that's the answer, right?

16:23Cause I've, I've had conversations in the last couple of days that have been very interesting. And somebody is like, Oh, prompt engineering. I'm like, I wouldn't call it engineering, but, but yeah, let's, let's, let's, let's, what do you want to talk about? And they were using a chat GPT in this case, and they just copied a little function that they wanted to translate to PySpark. It was a SQL one. And then just like, how can I do this in SQL? I'm like, no, no, no, hold up. What are you doing? It's like, well, I wanted to move this from SQL to PySpark. I'm like, but that's not how you ask. And I kind of explained that the, the, the, how to prompt you need to first tell it what it is. So I was like, you're a senior data engineer that is working on blah, blah, blah, like giving it a little bit of context and then explain what you're trying to do and then explain what are the best practices that you want to implement. And that's where that knowledge of software engineering, the knowledge of data governance, the knowledge of all these other things and other aspects that you start feeding the LLM and it gets enough context to give you the specific answer that you need.

17:26Right. So I think it's very important to have knowledge of superficial knowledge of how things work in general will make you a better prompt engineer. Right. Everybody's going to be a prompt engineer in the future. Like, I don't think it's a role. I think everybody should be a prompt engineer. Right. Because it's kind of like learning how to Google, right. It's the new learning how to Google. So I do think people should start looking into, into that. And it goes back to the curiosity piece, because I think curiosity is what builds that type of knowledge in us. If we weren't curious, we would not investigate and we would not learn. Right. So as long as you care about those things and you want to learn about them and you, heck, you can even ask the same AI, where can I learn this? Or can you teach me? Right. So knowing how to use those tools to also learn and research. And obviously there's going to be results that are not correct because we're not there yet. Right. But if you have critical thinking skills and take things for that chat GPT returns, and then use that to do your Google searches to kind of confirm and ensure that things are correct, that's going to make you successful too.

18:33So yes, I totally agree. People should start thinking of being more flexible and trying to learn new and different things that are around them so that they know what questions to ask and what to learn through the different means that we have nowadays.

18:48Sacha: So many things going through my mind right now, as we speak is actually three things. First one, as you said that, you know, people that can have a very superficial knowledge about a lot of things are going to be very successful, which I totally agree with. But it's so funny because there is this saying that says jacks of all trades, master of none, which I think may not be true. Well, do you know the whole quote? No, do you?

19:13Guest: Yeah. Well, let me, let me, I know that the quote is not just that.

19:17Sacha: Let's fact check that.

19:19Guest: Yeah. So it says jack of all trades is a master of none, but oftentimes better than a master of one.

19:26Sacha: So there you go. I agree with that full sentence.

19:29Guest: Right. So I was watching a YouTube video yesterday. It was, I also like to do 3D models with Blender. I have a 3D printer. I like to do that kind of stuff as well. And this person was talking about Adobe and Maya and all the different tools that they used to use and how being linked to a particular tool was detrimental because if that tool goes away or it changes, that person did not evolve to know things outside of that. You do under, you do learn techniques, everything. But imagine if you have to learn a different type of tool to be able to do, you know, nail down a nail. That is a very, there's going to be learning curves. There's going to be people that are not going to be able to do it. There's going to be people that are just going to give up and not do it again. So I do believe people should be generalist and specialists, if that makes any sense. It

20:27Sacha: does to me at least. And I think that's something that Slum is also pushing forward as well with the, the T-shape form consultant, where you can go deep in one topic, but as well, you know, wide on, on many topics. And this is something that I really agree with. And then, I don't know if you saw, but I think it was yesterday or maybe the day before that Google announced Alpha Evolved. I haven't seen that. And basically what they do, and I think it relates really well to what you just explained. They built a full AI agent that's able to, based on obviously a good prompt from a scientist or whoever wants to use that agent, go and solve very complex problems. And basically the way it does it is that you give it exactly as you mentioned, you give it what you have, and then you define the best practices that you want to see. You pose a problem that you want to solve. And then the agent is able to go and produce many, many different kind of potential response. And then another agent is going to look at all of these responses, select a group of potential high quality responses, and then go back and go in an iterative process over that.

21:37And they have- Yeah. And they're using the pro model

21:39Guest: and the other model, right? Which has different types of inference, I think. Exactly. Where one's faster than the other. So it's somebody, I know which one you're talking about. They were different types of algorithms too. Cause like the pro is more of like a deep search, right? Because it takes longer to do and infer and then, and what it needs to do versus the other model, which is more of like a fastest way towards the solution. Right? So this one that does all of these different solutions while the pro model just grabs the subset and then actually does the like thinking piece of checking each one and then doing that back and forth. Yes. I know which one you're talking about.

22:17Sacha: It's amazing. Cause I, I read a little about a little bit about the results and I saw that they have managed to complete and I forget the exact math problem it was, but it's a math problem that has been standing for over like 300 years and it managed to complete it. So really, really impressive. Now, you know, is it true? Is it not true? Like, I guess like we'll find out in the future, but very interesting. And then another thought that I had while you were, you were talking is to me, the, the barrier of entry for entrepreneurs now is so much lower. And I want to hear about your take on what does that mean for bigger companies that, you know, previously couldn't really be affected by that because now technology is becoming a commodity and a lot of people can now do a lot of things right at their fingertips. So I want to hear about your take on this. Yeah.

23:09Guest: I do think it enables entrepreneurs go faster, right? Like I've, I've done a couple of those. Uh, I'm not going to say what, but I've been basically it's not by coding. I I'm calling it like by businessy. So something like that, where you're asking a lot of questions of like, okay, if I need to do market research and understand how my market works and what's my market cap, it gives you enough information to do that. And then, okay, if I need to create a business plan, what do I need to do? And we're just like, it's a venture that I'm doing with my wife and we're both kind of going back and forth with chat GPT, trying to see, okay, what do we do next? How do we find this? And yeah, sometimes it's wrong, but like you go and find the information that you need and it's all fine. So I do believe it's going to enable more people now, just like if anything, once you make it easy, does it, is it going to have the same impact and value if it's that easy to do? So I think that's going to be one journey we're going to have as a species. And let's see what happens with that. Let's see how it affects the economy.

24:10Let's see how it affects everything. Right. Number two, regulation. So there's also a ceiling you're going to hit up to a point. Yes. You're going to be able to get maybe a good slide deck, a good pitch, et cetera, for, for your venture. But when you start getting into the real, you know, work and process of building a business and, you know, getting clients and all of that, that's where, where the fun begins. And, and I think leveraging Gen.AI as a tool is going to help enable those folks. But at the end of the day, there's, you're not going to know what to ask the LLM if you don't know what you're doing. So people like me that have been doing this for a while and have experienced a bunch of different projects, ups and downs, and seeing people do this with their servers and try to, you know, fix them with duct tape and bubble gum, you learn a lot of things and what to avoid. So I think in the future, as entrepreneurs and professionals, we should start creating our own kind of digital twin LLM type of deal where you come in like me as a professional, I could go to you like Sasha, I see you have a problem with this data and you need help.

25:24I have my expertise and I have my LLM with me that has all of my information as well, all of my notes and my, my things that we come together to solve your problem. Right. So it's kind of like your, your assistant. That's, that's basically what it is. It's your, like back in the day when we had guilds and you had like apprentice, it's kind of the same thing. Your apprentice is helping you out to do the tasks that maybe you don't want to put too much time or you're not in this mind space to do that, but they could do it and you just review it as the person that knows what they're doing. So the output becomes way better.

26:00Sacha: Again, so many thoughts and such an interesting conversation. I throw some of you, some of these at you and then you, you pick the ones you want to respond to. To me, you know, this lower barriers of entry is, is good for the broader society because that means that there is more choice, right? And there is potentially more applications that can, can resolve people's problems. Right. But it also means that we have to figure out which ones we want to use because now the market is flooded with solutions. Correct. Correct. So how, how do we do that first? That's my question.

26:32Guest: Just look at the video game industry, man. Like, no, seriously. Like if you look at how it corrects itself, I don't know how much you know about video game history, but Atari went through something like this. They were just creating games that were like the, the whole like Atari ET game. I don't know if you've heard about that thing where it was a disaster. The guy made it like in four days or some, some junk like that. It was a horrible game. There's a landfill in LA or California that where there's a bunch of them or Arizona or something like that. Anyway, the market was, was bad. Like all the output was really bad and people were getting pissed off, basically. So people start voting with their wallets. So yeah, you could have all these people creating these great things or not so great things and people are going to buy them. And then people are going to realize the quality of them and start identifying which ones are not good and which ones are good. And then it's just going to go through there. It's going to, regulations going to happen. Governments are going to step in or just the same people are going to by word of mouth, generate that, that funnel that sifts through all of the things that are not going to be good.

27:40Just look at all of the AI powered SaaS applications that are out there, right? There are one in a million, but there's a couple of them that are good. Like for example, Zuno, Zuno AI, which is a, one I've been playing around with to create music. It's great, right? Like it's awesome to do that, but there's other ones that you see in typical ads and either TikTok, Instagram, blah, blah, that you, Oh yeah, create your digital twin, blah, blah. And it's just a, on top of the LLM, not really like safe either. Yeah. So yeah, it's, it's, it's a natural progression of things. I've seen this many times.

28:16Sacha: I fully agree. And now my question is then what differentiates a good product from a bad product, because it's so replicable or more easily replicable, at least is it. And then you, you mentioned something that interest that was super interesting is, is it now me, Sasha, who becomes a business rather than my company, who is the business? Like, do I become a business myself because of my knowledge and experience, right? And this is the value that I can provide.

28:43Guest: That's, that's the way I see everything. You are a business always like, especially in our system, right? You, you, when you go to, as an employee, you are a business, you have a contract between a company and yourself. At the end of the day, you are operating as a business. So that's why when I see people like, Oh, you need to work on your personal brand. I do agree with that. Like I've been working on it since like five years ago. I have my webpage. I have, I did podcasting. I did YouTube. I do medium, sub stack, you name it. Like, yes, it is important to show your expertise. People are not going to know what you know, if you don't tell them, like that's number one. And yeah, you bring your, your knowledge and tools and understanding and experience to a company or to your business. Like that's, that's a given for me. Right. You did touch upon something else before I'm trying to remember what it was. Anyway, but yeah, that's my stance on, on, on that. What, what makes, what makes something good or bad or like worth your while I have, I'm old and I've, one thing I've, I've kind of figured out is beauty is in the eye of the beholder.

29:51Right. Let's take two IDEs. I know this one's like super, not controversial, but people have are very opinionated of this. Like what IDEs, you know, where you work, like a VS code or, or whatever, like a PyCharm. Oh no, VS code. No, something else or Vim. And everybody has their own like perceptions on what is good. So whatever the overall population thinks is good, that's what's going to survive. Right. And it also depends on the company and what they're providing. Some companies are super bad and they're still around just because they have one thing that people want and they keep putting money into it. Right. And there's some companies that are super good, but they don't do, they don't get the word out and not a lot of people get to utilize and be able to say, okay, this is something that, that has value. So it's very subjective, very subjective to understand what AI tools are good or bad. Like we're, it's your criteria. Right. I

30:45Sacha: agree. And, and, you know, like when I was asking you about, I, am I going to become like a business? I think it goes even further than that. I think in the AI age, if we can call it like that, I believe, and I strongly believe, and maybe this is very naive of me, but I do think that trust, even though it was important before is going to become even more important. And I think that if I look at myself, what makes people want to listen to my podcast, for example, would be that they understand and they, they trust me for to show unfiltered content and something that is true and genuine. Right. And then they, they know that they can trust me and rely on me for bringing valuable information. So I think that the trust factor is going to become increasingly important in businesses in the future. And I don't know what you think. And I'm curious to hear it.

31:36Guest: Yeah, I agree. And I think it's kind of remind me when you were talking a blockchain. So years ago when blockchain started bubbling up, I didn't see any like business application yet. And I didn't get into the space, so I don't know much. Okay. I know it's used for, for crypto. I know what, how it works. It's kind of like a ledger that is shared among people to kind of say, well, if I'm doing this, I know you're doing that. And that builds trust because there's no way to, there's a system around it. Kind of like laws. We trust that another human being is not going to harm you because there's laws that protect you that if they do that, they, there's consequences, right? So I think trust is going to happen when we do have that type of regulation in place, right? Where the, if there is a bad actor, if there is somebody that's going to do something malicious and there has to be a consequence for their actions. If not, there won't be any inherent trust where we have a signed contract or something that does back you up to do that. Right. I do not believe in blind trust, but that's just me.

32:39I don't think we should just take things for granted. Or if somebody tells you something is just because they say it is, I always am very critical of that. And I always say, Hey, you have to investigate. You have to know, you have to figure out what's behind all of this, right? For us to get to that level, I think we still have a couple of years to go for regulation and other rules to be set either social or, or judicial, you know, it depends on, on, on how we're going to take on that. But yeah, like we're, we're gonna have to figure out what the etiquette is going to be.

33:14Sacha: I love that. Now, if we go back a little bit into this whole conversation of definition of roles and what roles do what, et cetera, I have a question. There is a lot of CEOs. There are a lot of CEOs from major companies that are saying AI is going to just code everything, right? They're already saying that it's coding like about 30%, like that, you know, LLMs are coding about 30% of code right now in production, et cetera, et cetera. You're someone that's very technical. I'm curious to hear, should I still invest if I'm someone that wants to get in the field? Should I still invest in learning how to code? If yes, why? And do you see that being taken away from technical people anytime soon or at all?

33:59Guest: Coding is not just learning how to use a programming language and putting like writing, writing it out. And that's what the AI is doing. It knows the language and it writes it out. The important piece here is knowing the patterns, knowing best practices, knowing how things work, knowing how to create architectures, knowing how to create patterns in those architectures and having that way of thinking. That's what a developer is. That's what a programmer is. Programmer is not the language. A programmer is not the code. And if that is the programmer, that is in my opinion, okay, is not somebody that knows how to program. They know how to code, which is different, right? I'm going to give you an example. And I know I've used to, and maybe I'll keep using it, the interview question where I ask, do you know how a SQL clause is executed? And not a lot of people know how to answer that question. They think the machine just reads from the select all the way down, and that's not how it reads it.

35:01It goes to, it kind of hops around because it's trying to give you the smallest data set the fastest way possible, right? And you need to understand in the background how a database works and how it does things. Because if you know that, then you could do performance tuning on the lovely sequels I've seen out there, right? There's like nested sequels and this, and they do like crazy wares and calculations in the select. Well, okay, let's digest this and make this a little bit better, right? And if you know how things are executed, you know where to attack. But it's that. It's the knowledge of how to, how things work and what they're doing and what problem you're trying to solve, which is important. The coding, the tool, and all of that is, in my opinion, irrelevant, right? Tools are going to change. Programming languages are going to change. All of that is very dynamic. What isn't dynamic is how we solve problems. I mean, yes, because every problem is different, but there's techniques and patterns and best practices that we've brought with us during the years to kind of build these stuff, right?

36:02So, no, I don't think it's, yeah, learn how to code because you need to learn how it works. But I would invite most, or I would invite everybody to learn why and how it communicates in the back end, right? And what does it spin up? For example, Java is very different than Python. Like Java runs on a virtual machine, Python gets interpreted, right? It's knowing those things to say, okay, well, if I do Java, I'm adding all these other layers. But if I use Python and I run it, you know, directly into a headless server, I know it's going to run Python. That's how serverless works, right? Like it's a running server that just goes, grabs code and executes it, right? Versus a JVM where you have to spin up the virtual machine, interpret the code and do what you need to do, right? So, yeah, learning those things is important.

36:50Sacha: I love that. If a company is just starting their AI journey, who do you think they should and why? Do

36:59Guest: they have data, first of all?

37:00Sacha: Let's say they have data. Let's say it's unclean, maybe unfiltered. It's not in the right format, whatever. Like, let's say they have a bunch of data, but nothing is really structured.

37:11Guest: You just answered the question. Who do you need? Data engineer, right? Because you're building the foundation. That's what you bring the plumbers in first to do the base of the house, right? You don't bring the roofers, which are the data scientists and the ML folks, right? You're going to have to build a good foundation first to grab all that data, put it somewhere that's going to make a little bit of sense. And then you could start bringing in the rest, right? It's your foundation and you need to architect it well. I've encountered a couple of projects out there where they weren't architected well and they didn't think through a lot of it. And I see things where people think a data warehouse is just like a repository of data. That's not that, right? And they're tightly coupled with their business operations. And, you know, they have now it becomes an OLTP with an OLAP on top and it's just horrible, horrible, right? But if you have somebody that knows what they're doing and has done this before or knows how to do it, it's a better experience. You do have a solid foundation.

38:12You come in there, you say, okay, what data do you have? What possibilities? And you build a roadmap on top of that, right? If you do the analysis and you put in the time. If you just need to get up and go, you also need a data engineer and maybe a tool that will enable you to go faster. Think of those SaaS tools like a 5-time talent or any of those. Slap it up the credit card, get the thing, connect it to the data source, ingest it. And there you go. I

38:38Sacha: love that answer because I think a lot of businesses and maybe a lot of startups are maybe overlooking the data foundation and going straight away for AI transformation. But I think, you know, it's very hard to get there if you don't have a good data foundation. I

38:53Guest: think I should come up with the new AI version of this. But back in the day when we were doing big data and dashboarding was a thing, everybody was like, well, we just get Tableau or we just do Power BI or Click or whatever. And it's going to work. I'm like, no, all you're doing is grabbing garbage. And they're like, yeah, yeah, garbage in, garbage out. I'm like, no, buddy, you're going from garbage in to garbage. You know, and it's, you know, you're just slapping makeup on top of it. And it's still bad data, right? So data foundations are important and going in there and doing what's proper and coming up with, like, for example, data domains. I don't see a lot of people thinking about this, of how you're going to organize your data and what makes sense to you as a business. Are you going to go by department? Are you going to go by projects? Are you going to go by different forms of looking at data? Create boundaries around your data and those boundaries could overlap to create something new. And it gives you a visual representation of how things are going to work in your data platform and then how you can expose and democratize that data as well.

39:59Sacha: I want to start closing our conversation. And my question is always forward looking. So like, if we look into the future, I don't know, three, four or five years, because usually I would ask like 10 years, but now it's becoming so much harder to predict. What do you, what do you see as the next big thing for, for, for companies to focus on? Is it changing? Is it staying the same or where, where are we going, you know, in these, in these next years? It's

40:32Guest: always going to change. That's, that's for sure. That's taxes, death and changes are the things that happen in life. I suggest and believe companies should start looking into their internal data. Like that's an asset, both, uh, process and operational data. If you have that in your data platform and you start analyzing and you start creating your LLM and you start feeding that repository of knowledge, tapping into that potential of what you could get out of it. You could see how your operations are doing. You could create new insights that you just didn't know were there. You could enable your, your team to be nimbler, to ask a question, get an answer. You don't have to go and talk to so-and-so in this department and then go to somewhere else because that, oh, then that person left. You know, all of that should be centralized. Tribal knowledge, operational knowledge, all of that information fed into an LLM.

41:35We'll provide so much to companies that focusing on that in the next years, a couple of years, I think is going to be key to who's successful and who's not.

41:44Sacha: Are we going to see, this is my last question, I promise. Are we going to see a new type of, cause you know, we went from a data warehouse, data lakes, the lake house. Are we going to see LLM house or something like that? Could

41:55Guest: be, but I think it's, it's more of a, just intelligence as a whole, right? Because the lake house and all of those are, they're just patterns you follow to, to kind of organize data. Data could be data. It doesn't matter. LLM is going to ingest it and do something with it as long as we provide it. So I think it's more of a, we're moving away from the search bar. That's what, the way I see it. Not from, from those patterns, more from the search bar. Instead of going to Google, I go more into an LLM to ask, right? So that's going to be the way to kind of ask questions internally in any business.

42:31Sacha: I love it. Yeah. I do see agents on top of a lake house being a deadly, a deadly combination. I want to give you the floor and thank you so much for all of your insights. This was really, really cool. So I want to give you the floor to share any last words, anything that maybe I didn't ask and you felt like, you know, could be, could be shared. So just want to give you the floor for that.

42:50Guest: All I want to say is, is stay curious. I think if you're, you care about a topic, if you care about anything and you have passion, focus your energy into that. Even if it's not something like this, I feel we, we always tend to, or most of the people I've mentored, I hear a lot of like, well, what should I do? What's going to bring me money and this? I'm like, yeah, money is important. Don't get me wrong. But if you're doing something you love, even if the money is not quite there, right? Like of, of, of, yeah, I get, your needs are important, but if you're happy and you're doing something you love, I think that's worth more than, than, you know, the economic factor. So I always say, stay curious and follow your passion. That's, that's kind of my, my last words here.

43:32Sacha: Love it. Thank you so much, Aug.

43:34Guest: Alrighty. Anytime, Sasha. Take care. Ciao.