Energy Bytes

Derrick Turk, President at Terminus share his journey from mechanical engineering to leading a tech-driven company in the oil and gas sector. Derrick discusses the challenges of integrating data science as a core business function, the evolving role of AI in industrial markets, and the importance of domain expertise in tech startups. He also offers insights into the future of sustainable energy innovation and how technology will drive change in the sector.

Join the conversation shaping the future of energy.
Collide is the community where oil & gas professionals connect, share insights, and solve real-world problems together. No noise. No fluff. Just the discussions that move our industry forward.
Apply today at collide.io

00:00 - Intro
00:21 - Derek's Background in Energy
01:59 - How Derek Got Into Programming
06:53 - Derek's Experience at Devon Energy
10:02 - Tools Used at Devon Energy
15:18 - ArpsDCA Overview
20:00 - R vs Python for Data Analysis
22:05 - Interactive Development Techniques
26:35 - JSON Over HTTP in Data Science
27:50 - High Modernism and XML Explained
32:13 - Terminus Data Science Insights
37:32 - De-emphasizing Data Science Practices
40:40 - Understanding Data Revenue
42:50 - Favorite Projects in Energy Sector
44:59 - Highlighting Favorite Project
54:50 - Lightning Round Q&A
58:02 - Future of Energy Trends
01:01:31 - What Are You Bullish On in Energy
01:06:55 - Where to Find Derek Online

https://twitter.com/collide_ai
https://www.tiktok.com/@collide.io
https://www.instagram.com/collide.io/
https://www.facebook.com/collide.io/
https://www.linkedin.com/company/collideai

What is Energy Bytes?

Welcome to Energy Bytes with John Kalfayan and Bobby Neelon. Your essential guide to the intersection of data and energy. This podcast dives deep into the world of energy, shedding light on how data, AI, and technology are revolutionizing this sector. Each episode equips listeners with insights into the most efficient tools and resources, paving the way for a data-driven future in energy. From technical nuances to broader industry trends, Energy Bytes offers an unparalleled perspective on the evolution of the energy industry. Join us as we decode the algorithms of energy, one byte at a time.

All right, guys. We're back. We're here today myself, obviously, John Calian, my co-host, Bobby Nean. What's going on? Today, we're here with uh Derek Turks from Terminus. That's right. Founder, uh CEO, and uh head honcho over there. How's it going, man? It's it's great. It's good to be here. Appreciate coming. So, we'll kind of jump in like we normally do, you know, just give us a kind of brief background. When did you kind of get into uh into [clears throat] energy? Yeah. How did you you know, where'd you go to school? Yeah. What did you study? How how did that all kind of evolve to get you to the the Yeah. So, my my whole career path has really been the story of um slowly slowly kind of working my way back to what I always kind of wanted to do since like middle school or junior high. Um the the oil and gas thing was a little unexpected. So my my background the the the resume version is I'm a I'm actually a mechanical engineer by degree went yeah yeah went to University of Texas Hookham horns um but uh no I I kind of got mechanical engineering was a degree that you took because you didn't really know what you wanted to do uh the most diverse one yeah it was you know the pitch was you could kind of do anything but in practice what that meant is you were going to sit in a cubicle and design you know sprockets for Caterpillar or whoever um my dad's a petroleum engineer my My actually a lot of my family both sides of my family kind of have some oil and gas industry history. There's actually a cardboard cutout of one of my ancestors up in the um oh Titusville, Pennsylvania, the Drake well, the first oil well in the US. So you'll see Opa T there. That's a that's awesome ancestor of mine. So we we've got some history there. But highly highly recommend checking that out if you're ever in that area. It's not that far from Pittsburgh. It's worth the trip. Really cool. Yeah, percussion drilling. Think about that for a second. Holy cow. You can even take home some oil from the oldest producing well in the country. I've got some sitting on my desk. Yeah. But uh no, I I mean like I I I so I got into computer programming as like a probably honestly a little kid like fifth or sixth grade basic or Yeah. Writing writing visual basic macros in Word, right? Because it was and this is this is how so m you know there's a whole I won't go into this rant, but basically Microsoft kind of screwed everyone for lack of a better word when it came to learning programming. Um, a lot of the PCs in and kind of the prior generation, if you if you grew up 5 to 10 years before I did, um, you know, you might have like a Commodore 64 and it would come with basic pre-installed. It would boot into a programming environment. Windows was a lot more of kind of a a consumer device, right? Um, granted with 20 years hindsight, it's like nothing compared to iPads and and phones, [laughter] but but you know, still it was it was fundamentally not designed to to make you nudge you in the direction of being a programmer. And so, you know, as a kid, I kind of discovered this thing and I was like, "Oh my god, I can like you can have a button and when you click it, it's like guess the number." Yeah. Just really cool, really awesome games, right? I didn't have a Nintendo, so that was my Anyway, cheap entertainment. No, man. That's um And so, you know, I kind of being in high school in the midst of I guess the the.com boom followed by the dot bust. Um all the advice I got was like, "Oh, no, don't study computer science. That's a [laughter] that's a dead end. You know, that'll be that'll be over in in five years. And uh so I you know anyway took mechanical engineering as kind of I didn't really know what the heck to do. Spent the entire degree trying to find ways to work kind of programming into that. And what at the time we you know a AI is this funny word, right? Because it's just always whatever is like five years out. So at the time AI meant stuff like genetic algorithms and these wacky kind of global optimization techniques that anyway so we played with a bunch of that stuff as kind of a a hobbyist all through through college and um ended up kind of you know my given the family history had a lot of people trying to persuade me that the the oil and gas industry or I guess what we now have to call the energy sector um is a good place to be and uh took some internships kind of enjoyed it a lot of interesting stuff going on a heck of a lot more than what was going on in the more conventional mechanical engineering world because you I mean it really the comparison was like you were going to sit in a cube and be responsible for sprockets on this particular line of of tractors versus you know and I kind of was was strongly drawn to reservoir engineering where where you sort of see this whole picture of right these you know multi-million dollar operations start to finish and the entire life cycle these big capital projects important yeah and and it feels like you're actually fundamentally producing you know I hate to say it but my my wife uh Katherine also a mechanical engineer by background also did something totally different. Um, went into the chemical industry. We talk about making molecules, right? Like there's a and it it sounds weird because I'm I'm on a tech podcast and obviously we're kind of in a tech role. Um, but I just I love the draw of working with people or working in industries that actually produce tangible Yeah. tangible stuff, right? We don't we don't do a lot of that anymore in the US. That's that's as deep as I'll go down that rabbit hole, but that's No, that's there's a draw there for me. It's true. I I didn't take my PE exam because I was like, I'm not going to go I don't want to go do that. I don't want to go design parts and then if they fail, right, be out of a job for the rest of my life. I also didn't trust myself uh mathematically. Oh my god. That's a lot of stress. Differential equations pretty much ruined any dreams I had of becoming a [laughter] uh getting a PhD in Mechy or becoming a Yeah, that was still remember the spring equation from Diffy. Yeah, they're really only thing I remember from it though. [laughter] Oh god. Yeah. Hook's law or Well, no, not even Hooks. Yeah. Yeah. Yeah. Having flashbacks there. No. So, yeah. Like, no joke, one thing that really drew me to programming is all the math is like there's no real numbers. Yeah. It's all integers. It's all discrete math. If there there's like an alternate world where they taught kids discrete math in in middle school and I would have loved it like that. Just would have made [laughter] math versus geometry, right? Or and calculus. But, oh well, it you know, it is what it is. Um, so those of us who who can't do the calculus are kind of drawn to Yeah. integers. Yeah. It's f it's funny because it's like it's one or the other. I feel like generally, right? Like I'm a very visual person and I'm also very I'm too logical at times and so like if it's theoretical math where it's like prove this proof. Yeah. Why? Someone else already did it. I don't care. Like Yeah. And that was basically my college experience. I mean that's [ __ ] That's what half of Cal and me in college is is just proofs, right? And derivations, just endless derivations in and so it's like I was a C calculus student, but I was an A physics student using the same calculus, but it's just applied. And it's just it's yeah, I I I remember getting a kind of handwritten red ink note on one of my fluid mechanics exam that basically said, well, not basic, literally said, you do not understand the principles of Bernoli's equation. [laughter] And I was like, yeah, you're right. That's right. I don't understand this at all. Um, so it was, you know, it was educational, but uh, long story short, I I kind of almost immediately discarded a degree other than as a, you know, kind of line item on a resume and, uh, ended up going to work for for Devon Energy. Okay. First in Houston and then, uh, up in Oklahoma City later and um, kind of found my way into being a a reservoir engineer in their their New Ventures group, right? So you were in the New Ventures. Yeah. Yeah. And so what and so new ventures, you know, this was like 20 2009, 2010, 2011. Devon was a little little behind the curve on like Eagle Ford and in oil shell or unconventional oil in general. Um, and so we were we were trying to make sense of this stuff any way we could. Um, including through reservoir simulation, which really you think about some models of town where you had basically no data. you kind of build this three layer cake and say, "Oh, yeah. This is a, you know, under this massive set of error bars, here's the massive set of [laughter] error bars you get out and somewhere in here." Yeah. It's kind of a totally fruitless exercise. So, what what really started to to take off is is using more statistical approaches where we we we started saying, you know, okay, I I have no idea what the actual mechanics of this rock are, what the for that matter even what the what the properties are, right? There's a whole, you know, what is permeability in a shell? Well, let's see. We uh we drilled this core from 10,000 ft in South Texas. We took it up to the surface. We put it on a truck. We shipped it to Colorado. They crushed it up. Uh they put a transient pulse of helium through it and measured the and differential equations comes in, right? So there's some fancy math with differential equations and that turns into some number that they call system permeability. You're like, I don't know what this has to do with anything. I can plug it in my model. But so anyway, we started taking the approach of, well, here's what I can measure, whatever that is, right? whatever variables we had, here's what I care about, right? Which is like IP30 or 69, what whatever some short-term production data you had, and you just started looking for for correlations, right? throwing whether it was predictive models or just sort of u you know things like clustering algorithms or or any way to look for for trends and patterns or or even very modest kind of leverage on the problem of especially as you got farther down the the life cycle into like frack optimization you know we may not understand any of the mechanics going on here but man we see that there's a you know we we can get like a never trust any R squared above point8 [laughter] we can get like a 72 um on this model and if we kind of you know hold everything else constant and lock in the average values for this reservoir and we you know we draw the response curve of like IP60 or or IP90 or even six-month fume if you live in paradise and have that um I can draw the response curve of that versus profit loading and hey look diminishing returns or some kind of scurve and we want to sit kind of like right here at the elbow. Yep. So we, you know, we started doing stuff like that and it it yeah, I was actually kind of the first staff engineer in Devon's. Um I don't think they call it data science. I I don't know what they call it now, but basically data science today. It it grew out of reservoir engineering for the the the reservoir reservoir engineering technology group that sort of supported new ventures is how all those initiatives started at D. Interesting. And and if we want to kind of tie this in like history lesson of what tools have you used and so like what were you all using at the time to do some of that stuff? Oh man, this is one of those that's going to get me in trouble cuz this is like the why did I go? Well, all right. So, there's there's a great episode of Always Sony of Philadelphia where they like they put on a musical and and [laughter] this question keeps coming up of like who are you but who are we doing it versus? Right. Yeah. So, I'm that kind of person. I've always got to be maybe it's weird, but my my motivation is typically like a all right, but who am I doing this versus? Like, what's really kind of grinding my gears lately? So, I'm not going to I'm not going to name names, but we a large three-letter acronym statistical software vendor um that I dealt with over the years um provided very expensive tools uh very expensive consulting around those tools and um to be honest, we were kicking the crap out of their results with free software with at the time things like R. Um honestly, a lot of the R ecosystem, little bit of I mean, my god, what was it 2008? there's probably Pearl scripts floating around, a little bit of C++, Python, kind of a grabag of everything, [clears throat] right? But um you know, I was just kind of this shocking experience of seeing how industry was being, you know, it's our industry is a little bit conservative for better or worse. And so we we we tend to be drawn to the big vendors, the you know, nobody ever got fired for choosing IBM type approaches. 100%. Um, but man, it was it was brutal seeing like how much the and not just at at the company I was at, but across the board, how much companies were spending on some of these uh big proprietary software packages and then just like the quality of the output and the analysis versus what you could do for free, right? Literally piggybacking off, you know, obviously years of work by a lot of people. Yeah. But um yeah, it was incredible. And and so you know that one of the big factors in in kind of realizing in 2013 I could I could kind of jump out on my own. And there was this need in the industry for you know the logic at the time is there were obviously a lot fewer people who could do um kind of what we call data science or or analytics or whatever the you know the buzz word was. It changes um and had the domain knowledge of oil and gas versus even what the the big you know uh software vendors or IT consultancies could offer. I mean it just there there was such a shortage and of course this was I think I've got a slide in in a talk I've given a couple times that shows like the Google uh engrams trends for like data science versus time and of course 2012 it just like shoots up because it was um oh Harvard Business Review Times had the whole like sex right there was this article about um I think they called it analytics or whatever but Tom Davenport and then Moneyball blew up and you had all this all these things kind of kicked in at the same time and it it suddenly became like all anyone was talking about. Yeah. Sexiest job of the 21st century. Um and then that guy went on to actually become the US's the the first official like US national data right under Obama. It was his whole PR. I mean anyway so it obviously the timing was right and I kind of jumped out and you know uh I was kind of liken it to like building the engine as the plane is in flight. Yeah. [laughter] But that's a very good analogy. So that was Devon into just on your own like jumped out on my own and um went went out independent. It it was just me for uh many years and and my wife uh Katherine actually joined the company in 20 well this is terrible now I don't remember couple years 21 yeah it's been about two or three years um again coming from the the chemical industry and kind of having some experience in things like predictive maintenance and there's a whole other world of like um things we can learn from in our industry that people been doing in in these kind of very companies that we tend to think of as kind of boring or conventional but you look at your well I mean lean manufacturing, right? Literally the simplest [ __ ] right? Like I was doing this as an intern in college and when I was an intern, it was 30 years old. Yeah. You know, and then I'll never forget when I saw the uh article in the uh chronicle, sorry, um about Hess announcing that they were implementing lean manufacturing methods in like the mid whatever 2015, 2016, something like that. I was like, "This is 40 years old, guys." [laughter] Like, "Why is this front page worthy news? This is not a new thing. I've been in organizations where, you know, oh, there's a six sigma, a new exciting thing that, you know, like you said, been done since Toyota's been doing forever." Forever. Yeah. Yeah. We're finally kind of catching up with the hottest trends of the 80s. Yeah. [laughter] You know, what's going on in Japan is really interesting. We really need to look into that. Um, [clears throat] like agile and all I mean it's like agile manifesto was however long ago and but then we could probably even get into a whole how it's probably not the best workflow for data type projects. Yeah. [clears throat] No, it's well that's a yeah there's a there's a whole conversation about what what is what even is agile [laughter] if you look doing scrum every day right? Yeah. Exactly right. Like any uh well it's a whole maybe I won't I won't say that on air [snorts] but there's like a lot of movements in history when you look at what the original founders intended versus kind of what it got into like oh my god. Uh, [clears throat] very true. Yeah. So, I mean, one thing I want wanted to kind of hit on, and people may or may not be aware that this was you, but there's ARBs DCA package. Was that at Devon? Was that once you broke branched out? Like, um, that was something. And, to be honest, I I don't even remember what I I think that spun off from a a project we did where we kind of, you know, ended up at having done a bunch of analysis. We, you know, we we by necessity built a um an R library for for fitting ARPs declines and said like, well, this would be a cool thing to open source and and kind of use as both a well number one, so people stopped reinventing the wheel, right? And because it so a little bit of boring history, but at the time, Spotfire was probably still is, but was kind of the BI tool that was had taken a hold in the the oil and gas industry. Spotfire went for a long time had had no real programmatic interface beyond iron python which you could actually sort of we we had tools in Devon and what S+ well not not even at that point I think they had an external interface to it but it it was pretty obscure like okay yeah I started with Spotfire 2014. Okay. Yeah. So like at that point they had S+ and I think tear was on its infancy but yeah so at Devon we had um I I remember this project we had a I I built a a random forest model for frack optimization like in the I think it was in the Barnet shell and what I ended up actually having to build is kind of a I'm going to say for lack of a better word a a compiler that would translate that generated random forest model. In other words, those those tree kind of the the the decision [clears throat] tree, right? The kind of if ends in in the random forest to um two iron python code that we would then run in Spotfire so we could actually embed the model. And of course now, you know, they they've got tipco for whatever reason decided the the real way to to uh open up custom code inside Spotfire was to sort of white room implement our odd choice given it's already open source, but whatever. I think they wanted to own it and so that that was terror which is kind of this almost compatible reimplementation of R and and now I you know you so right so you can write our code to implement sort of custom calculations inside spotfire. Yeah. Um and that took off about the time we released ours DCA because people started asking like oh cool like if we have our code to do this we can run it in Spotfire. And so all right well put that out and that's kind of what made that that blow up. Um, and yeah, people, it it's really funny like I've had the experience multiple times um of getting an email, you know, in my capacity of of as Terminus, let's say, that's like, hey, we've been using this uh this R package called ARPS DCA and we're we're looking for someone to help us integrate it with Spotfire. [laughter] And then every once in a while just like look at the author Yeah, the [laughter] author tag on. Yeah, we can help. Might be able to help you with that. I think I'm fair fairly familiar with that chunk of code. Yeah, it's funny how uh some of that stuff wraps back around when you when you publish code as as open source. It's funny. Are are there still uh commits and stuff to it? Uh you know, yeah, every once in a long while I'll I'll take a bug report or something. It's I I don't think we've made meaningful changes to it in years and years. The equations haven't changed. Yeah, the equations haven't changed. I'm not saying it's kind of perfect or flawless, but it's, you know, it seems to do what it's intended to. And we we don't do a lot of our work these days. I I don't think the industry I like it all comes back like in our industry, it all comes back to Spotfire. So, as soon as Spotfire opened up the ability to write data functions in Python, that kind of you know, R has lost its luster a little [clears throat] bit. Yeah. And I think we were kind talking about that the other day on the the upstream Slack channel just even about how R and then now there's Julia, but it's a matter of like, well, you know, is it too late to the party? And you know there's there there are better options but I mean like I I personally love R. I mean as far as like I I would like to use it more but then there's still this whole can't use R in production thing and there are there are limitations to it. I mean it's I I have such mixed feelings because it's a like it it's an interesting specimen as a programming language. It's not a good programming language but it's a it's a really interesting one when you put it in context of sort of like what is it like what is it really? And that's probably a That's probably for a different podcast, but it's it's it's a living fossil. It's a it's actually kind of in the lisp family of languages. And it it's actually obscure even within that because of the way it implements um well Hadley Wickham would call it lazy evaluation, but what it what it really is is like this other thing called F experts, which is there was like one researcher that was into this and he's dead so it doesn't anyway [laughter] like I said topic for a different podcast. um it's kind of a cila can't but you know the thing that R so I like I I'm I'm a I'm I am also a believer in like don't use R in production and and for me that that comes from using Ruction in [laughter] production um and just testing suites and yeah you know it's not like Python's great either but it's more robust got plenty of gripes on Python but there there's also kind of the it all comes back to fashion and network effects right and it's like there there's safety and numbers um the thing I miss from R is the the graphics, right? The plotting. Yeah. We we were we mean my wife and I were talking about this the other day as one does at Terminus HQ on just how there's there's in our opinion there's really no nice easy to use um sort of JavaScript or or web client side graphics library um that produces good output out of the box. Right. And I I know people write in what they're like, "Oh, hi charts plotly." Yeah, yeah, yeah, whatever. I know. But um I but but you got to remember I'm comparing to like ggplot 2 and graphics and just like I don't think people who didn't use that understand how easy it was to just say like I've got this data frame. Hey show me um you know EUR versus lateral link but break it out in panels by uh landing zone. Yeah. And and it was a oneliner right there was kind of this domain specific language for have all the setup for the plot and all. I think there's someone trying to implement grammar graphics for Python. I have to find the think that'd be great. But I mean because one thing I I liked you talk about like for the web you know client side ver you know I loved in R then the plotly package you could do ggplotly and just basically plop your ggplot code inside of that wrapper and it would just turn that into the interactive you know web version too. So, yep. So, I don't know. We were big D3 fans and I I think the I think the monetization story there with um what are they now? Observable HQ kind of the notebook uh they'll they'll they'll let you use D3, but they really really want to sell you a subscription to their cloud-based interactive notebook thing. And it's it's made the the documentation just completely unusable. Um in my opinion, I I'm kind of in the same boat with you. My favorite thing about R versus Python, generally speaking, is the R Studio IDE. Yeah, the variable explorer and the plot. Like, as an engineer, I like to and just generally speaking, as a visual person, I like to be able to see, okay, as the code runs, there's my variables come popping in and what is the value of said variable instead of having to put print into my Python statements. I think we talked about this before, too, like the ripple is like a huge deal. And that's why notebooks won't die. That's what I mean like there's right and and and rapples are so interesting to me because like so R all right we'll go into a little yeah let's go into R is is literally it is kind of descended from the lisp family of programming languages and so a lot of you know if if you kind of don't know what we're on about there's there's this idea called a ripple read evaluate print loop or or an interactive interpreter and it it's kind of the thing you can do in in Python or R um or I suppose JavaScript if you drop into just like node from the command line where you can enter one line at a time or one one statement at a time if we if we want to get real technical and and have it evaluated and and sort of you know see the result. Um and if you're coming from something like Java or C++ you know obviously it's like oh my god this is interactive this is really cool I don't have this compile step in my in my loop. Um but I think people have you know the generation that maybe started with Python or JavaScript doesn't really know the limits to which you can push that. So, lisp, the lisp family of languages, um, in in and small talk I I would put in kind of a related category really played with the limits of what you can do with with interactive development. And so, like if I'm in a common lisp environment, and I have been once or twice in my life, um, sort of like you said with our studio, it's not just, oh, I'm entering a line and seeing the result. It's like okay, I've got my pro my source text over here and I've got kind of a running uh you know ripple over here and then maybe I've okay I see this variable in my source text and I'm going to hit a keystroke and that's going to pop open an inspector and I can see oh it's an object of this class. Okay, it's got these elements. Well, let's zoom in on that. Show me what the um you know what the monthly production field of that is. Oh, okay. That's an array of floating point numbers. Oh, [snorts] oops. That one's wrong. Let me edit that. Okay, I'm going to call a function on it. that crashed. Oops. Let me uh let me go and edit the code of that function. And so it's kind of, you know, this environment where everything could be um sort of explored almost like hypertext, right? Or like a like a website where you could navigate between uh data elements in the running program and um you know sort of interactively edit source text. And it's just a it's a totally different way of working, right? like uh even the Python ripple or notebook is still very much a uh it's it's like the the butterfly that's been kind of pinned to the the board and and preserved, right? It's you can you can see it. You can kind of play with it, but you're not seeing it in action. You don't really have you you've got um oh, you've got kind of static snapshots at points in time, but you don't really have the live object. And so that's you know, yeah, I I completely agree. [clears throat] people who haven't worked in an environment like that. Um, it's probably like VS Code though now because [clears throat] you can run your your Iron Python notebooks in VS Code, but then you can also have like the variable explorer. So, yeah. Yeah. Um, so there's Yeah, I I would love to see more hooks and maybe the Jupiter uh people will will kind of build in that direction. Um, there's no reason you couldn't on top of Python, you know, it's the the interpreters got all the hooks to kind of introspect objects at runtime. Yeah. But uh well, it's just like I feel like with some of those languages, right, you've got people like us who weren't necessarily trained in coding but are very logical and understand logic and math and like coding. And so for someone like myself being able to go through essentially line by line and kind of debug and see how the logic works and see what is in which variable at what time and figure out why it's not working is a very helpful thing. Yeah. And and I like the ripp pools from like if I'm especially if I'm working on something and I don't know what I want yet, but I'm getting it from an API, right? Like I don't want to hammer that API every time I need to run the code like or save it to a JSON file and then write some sub code to go, you know, I can just store that JSON object, you know, or even a data frame of that or whatever and then I can iterate on just on that object from that step rather than having to repull it every time. Like, so I think our industry is actually really lucky in that we're a little bit slow to adopt things. Um, I say this, [clears throat] this is this is probably me just rambling about a current project. Um, for those of you who've never had to deal with this, how fortunate we are that our industry didn't really adopt web APIs until things have become predominantly JSON based. I'm going to put I'm going to put rest in massive scare quotes because there's this whole holy war over Roy Fielding's thesis and what really is REST and basically my cards on the table. I think the original definition of REST is kind of dumb and useless. But like if you treat it as RPC over JS like RPC via JSON payloads over HTTP, like I think that works great, which is what we do anyway. Anyway, so the real thing is had we gotten started 10 years earlier, everything would be freaking XML and SOAP. And I'm I'm working with a a financial client, a financial industry client right now, and it's like, you know, they were earlier adopters, like, oh my god. Yeah. Um I I think technologies rise and fall basically proportional to um how hard is it to figure out what is going on when all you have is a crash dump and a text editor. [laughter] And and with JSON, it's a heck of a lot easier than it is with SOAP. So that's kind of my theory. Yeah, I've had to touch a couple things at different points and it's like why like This is awful. Um, it's funny because it's like the concepts behind some of them are really good, right? Like with ML having structure and naming conventions and it's always in this format and this structure, but then it's like, yeah, yeah, you get into people and they're like, no, it has to be this. Well, I could just replicate the database to your database and you actually have way more. You have the same data in in a structured form where you don't have to write a bunch of code to ETL it the way you want it. But yeah, I I think there's a whole throughine between um let's see some some some odd touch points I think kind of triangulate what All right. Um it's high modernism. There there's this guy James C. Scott. He's a I guess a anthropologist, historian, anarchist, a little bit of all three. Um he's got this book called Seeing Like a State and it it's kind of this this is I promise there's a point to this. It's kind of this critique of how in the 20th century um actually not not just the 20th century, but it it took off in the 20th century. um the world saw this this kind of um wide spectrum of programs and and policies that took the approach of like we're going to have the perfect design top down um modernism right high modernism uh like corbusier designing uh cities that were these perfect geometric grids or uh the Soviet Union coming in and collectivizing farms and trying to put a rational five-year plan on them. And I think there's just kind of this um you know there's this human hubris we have of sorry this is this is I view XML in the same XML is like the the winl thing is the same category to me. There there's there's also a Richard Gabriel essay that has to do with Liz versus C called worse is better. And I the punch line of all this is like often in in some field of human endeavor there's kind of two approaches. One is smart guys sit in a room, think of all the cases, and write down the perfect flawless shining crystallin gem of a plan that's going to compass everything. And the other plan is like let's just let people do their thing and see what happens, right? Let let whether you call it evolution or the free market or you know, whatever you want to call it, but this idea of let's just let things kind of compete naturally work out. And you know, systems designed along the second and a design may even be a misnomer. um systems that come about through that kind of organic evolution seem to last a little longer, be a little bit more robust. I mean, you look at all the open source stuff, all the languages, Bluetooth, Wi-Fi, like they're still mass like built by ma, right? The the other classic essay would be um the Cathedral and the Bizaar, right? Eric Raymond, I think in the 90s, had this kind of take on um, you know, the development of of Linux versus kind of proprietary software and this idea that, you know, maybe for some things it makes sense to have brilliant people sit in a room and and build their, you know, they're flawless architecture, but for the kinds of things that have to interact with a lot of different edge cases, you you do a little bit, but then you have it may be flawless, but that is flawless based on their biases. Is that correct? Yeah. And it's it's this thing that's perfectly right crystallized to their well I mean their biases what they know their you know what they accounted for right right it's uh going back to list keeps coming up I guess it's kind of a a cultural touch point in programming but um you know common lisp but it it locked in all the best ideas of the mid 80s so you you've got this um vast kind of standard for everything that should be in uh the standard library and how things should work and it it touches on oh all these concerns that were really pressing at the time like you know the different is my character set ASI or EPODC or whatever and then of course it was like but what you know parallelism concurrency like that's not in there because right it wasn't a thing you know there's [laughter] no yeah I think I think that it's a good point have to be able to evolve if they want to have sustainability they can't just be completely locked down without any room to grow I mean that's what you with all the big big companies, right? Is it's like they round up a bunch of really smart people and they put all these handcuffs and constraints on them and bureaucracy and it you know permissions and all this stuff and then they can't really grow it and evolve it like it should be if it was just open to to them. [snorts] So kind of moving back a little more towards Terminus like I mean so is it mainly like a lot of projects or do you guys have any products like So we're we're always we're like constantly one year away from a a product. Yeah. [laughter] Um and we've we've looked at a few over the years. We you know the bulk of our work is is consulting and we've kind of I guess the interesting story is when we started we were Terminus data science LLC and of course data science was you know sexiest job of the 21st century and it if if anyone remembers that time there were all these great ven diagrams about uh data scientist being a a unicorn right you had to have I think one was uh domain knowledge and statistical and then programming or computer science and then the intersection was data science science. I thought that's great. That's exactly what we do. You know, that's that's so true. You know, if if someone's got that intersection of skills or an organization does, they can really come in. They can make a big impact relative to their resources. You know, you kind of this this high leverage um kind of approach. And of course, I I don't you know, I've got a talk out there somewhere I've given variations on a couple times called like Save the Unicorns. I think we did this at at Saga Wisdom U Sofa Becker last year. And the the premise is that somewhere along the lines, we we got kind of caught up in in trying to be Google or or Facebook or emulate what was coming down from these massive Silicon Valley uh tech companies and and we sort of let that role of data science become this super specialized siloed thing. Um I I'm going to rag on data science because I I see this on LinkedIn all the time. I've probably seen three posts in the last week that boil down to like companies need to invest in data pipelines or data engineering because otherwise the data scientists just don't have clean data to work with. And it's like, guy, I don't know what you think your job is, but that's what um I'm sorry if if your role is to take pristine, normalized, uh, you know, flawless, well-defined data and plug it through, I guess, AutoML and then say, "Hey, boss, the R squ is 72." Yeah. Well, I can do that. Like, I mean, I'm I'm not a big um I'm I'm kind of an AI skeptic. I I don't think Chat GPT is going to take our jobs anytime soon, but that guy I don't know. It may take his job. That's not much of a to to quote the the consultants in office space. I mean, what is it you would say you do here? [laughter] Um because I think that thing we call data engineering is actually closer to the the sort of the spirit of what data science originally was or at least a big piece of it. Um you know, we we kind of have this tendency to want to call things data cleaning or or I always, you know, in in programming we call it glue code, right? which is I suppose our fancy I would say it's ETL or maybe ELT ELT now right on how I [laughter] someone will eventually clobber me over the head hard enough I understand the difference there um but it you know are arranged in a different way right yeah so [laughter] it right so they're different but fundamentally you know I I don't know I I maybe I have this this um archaic belief that a a data scientist or let's just say a a um a person in that that kind of hybrid role of domain expert with some statistical and and software chops, it shouldn't be that big a deal to uh you know say recode missing values in a table or uh write a little bit of SQL to join two sources or it's not um I I'm trying to belittle it. It's actually it's um it's foundational, right? it really if you can't get that right I I sort of question I think that's the I think that's where a lot of people have the problem right is the sexy thing is is the end result 10% of the job but yes I completely agree with you the minute you start going down that rabbit hole even as a hobbyist right like oh [ __ ] I've got this API from this JSON now what or you know like looping through JSON getting back [clears throat] to that ven diagram like I mean this is supposed to be the person that has the domain expertise so you should be able to look at that data and say like this data is not right like or like or I should be these are these are bad values I shouldn't kick out like I mean like yeah for for all I complain about like oh they can't write SQL which is a valid complaint but what's scarier is is kind of taking the domain expertise out of the loop I I see data science becoming treated more and more like an IT function and it's like data comes over the fence I you know I won't name names but it's like I have been in conversations where the data science team have been asked to uh forecast production and so they you know went down all this rabbit hole of time series models and ARMA and Arch versus Garch versus And it's like, okay, but have you looked at ARPS? What's ARPS? It's like, let me give you a citation from 1945. Yeah. Um, and I'm not saying ARPS is the the end all at BLF forecasting. What I'm saying is, you know, if you're coming at it from the angle of ARPS doesn't adequately capture the, you know, the transient behavior of unconventionals. Yeah, I totally agree. As a reservoir engineer, if you're coming at it from an angle of um, I'm going to take time series data. you know, it's just a generic time series and it it could be widgets sold per quarter or it could be well maybe not. You know, there's a there's a different set of physical phenomena behind it's not seasonal. It's not like right, you know, depletion drive is a little bit different than like seasonal trends and babies being born and how many diapers people buy as a consequence. Um different different mechanism, right? So we we've kind of deemphasized the data science and our branding um just because you know we're kind of not the primadonna that's going to come in and say I don't have so and and and we've also discovered over the years that there's a lot of need for um automation and custom software and you know I I always tell people because there's this whole argument of build versus buy and there's no right answer but what I do tell people and it it's a little bit self- serving but it it's also true or I believe it's true is that if you are say you an EMP or you're an engineering consultancy or you're a, you know, a trading shop or a PE, whatever you are. Um, if you've, you know, you've got some process you've designed and if you really believe there's proprietary advantage in that and and you should, right? I mean, that's kind of your thesis if you think about it. Um, there's pieces of that it does make sense to build versus buy. You know, you're not going to get that off the shelf. So a lot of what we do is come in and help find that, you know, maybe five or 10% of the overall process that really is core and and special and proprietary and turn that into sort of robust software that makes it a repeatable automated production ready process. Yeah. Yeah. Yeah, I think that's that's a big thing for me too that you know just getting into this over the years as a started out more as a hobby but then became my job to a certain extent you start to realize right it's like you get so excited the first time you run whatever script or model or whatever locally on your in your IDE or in R or whatever and you're like okay well [ __ ] how do I get this so I can have it customerf facing like oh that's a lot different than it's hard right now you got spin back architecture you've got and it's and it's gotten UIUX the the more that we sort of chase the um you know kind of what's coming from Silicon Valley I mean it

Google has a certain scale they operate at and they have a certain base of talent that they draw from and they have I mean literally just an an army right like there are Google has more programmers I'm now I'm like thinking about this thing about to my like Persian Gulf War force stats. I'm like, oh, I don't know what rank they would be on the list of like largest armed forces in the world, but [laughter] if you armed every programmer at Google, they would be in the top 10, right? Like they would they would compare favorably to a lot of countries in the world just to headcount. And and so, you know, the kind of process that makes sense when you need to make an organization with literally, you know, tens of thousands of people um sort of function in a in a plug-and-play way. um very different. There's trade-offs they're going to make that don't make sense when it's, you know, 10 guys in a server in the closet. [laughter] And that's not that's not to knock the 10 guys in a server in the closet. That's actually a lot of my favorite clients are kind of at that scale. Um but probably launching like a Kubernetes cluster on AWS is not going to be the cost effective way to help them achieve their goals, right? And so it's Yeah, exactly. So there's um you know there's just kind of like a scale awareness that we've got to we we've got to build in the industry. Yeah. So when you know what are some if you can however much you can speak to it but what are some of your favorite projects you've done and then also what are you guys using? I know you've sort of using some rust lately but is it a lot of Python? I mean what tools databases. So we yeah we do a little bit of everything. My my like general advice on databases is you can for I'm going to hang some asterisks on this. I'm a reservoir engineer. I work in a lot most of my clients are dealing with um maybe production data, maybe uh daily or monthly financial data. What I'm really getting at is we're not we're not dealing with like 3D seismic. Okay, I'll I'll caveat it. you don't have some of the super high volume geoscience stuff but man in our industry like it is it is shocking to come across a workload you can't handle with Postgress and I I love Postgress I mean I you know obviously we've got a lot of Microsoft heritage in the oil and gas industry so you find a lot of like SQL server and that's fine um but man SQL server with with it you know kind of geospatial stack or postgress with postgis covers so many of the needs of of what we have in the industry and you can people will another fun like ETL versus ELT The other one I can never remember is OLOP versus OLTP, right? And in my head it turns into like cuz I mean it this but I translate it to like row oriented versus column oriented because that's what I really that's what people use it to mean in practice. Um but you you know for most of our workloads like it it's a distinction that doesn't really matter a lot of what we do. Well that's one of the things I'm learning as someone who hasn't ever had to set up any kind of backend infrastructure. It's like uh use the tool that makes the most sense. Yeah. From a cost and functionality standpoint that you've got at the time. Hopefully, you can also try and forecast out what this will be. Yeah. A couple years. But yeah, there's also a chance that there might be something that comes along that's better and cheaper or Yeah. designable. That's I was just going to say like it's you you can't plan you you don't know what's coming, but you you can plan for how do I build this in such a way that I'm not hooked to super specific features cloud vendor this database. Yeah, exactly. And that's a that's a huge chunk of so our kind of talk about favorite projects like we're we're in an odd niche and that a lot of our clients do not even have necessarily an IT department. We've got a lot of clients that are small EMPs or uh maybe small consulting firms, small might might be like a you know a PE shop or or a non-op. I assume that makes it easier sometimes you know it yes and no, right? So, you know, kind of like talking about the the the modern data scientist and how it's very much like, oh, please spoon feed me like perfect data. I I think those guys would come in and be totally lost. Um, but for us, we, you know, we find a lot of our clients don't necessarily have they they may have it one or two people in house or maybe a contract shop that helps them like, you know, manage kind of the traditional IT stuff of like keep the email server going and you know, Windows log on. Um, but a lot of what we do is green field, but it also is green field in a way that like we know we've got to leave something that requires minimum maintenance. Yeah. And so what's funny is it actually nudges us away from a lot of what's popular right now. If you look at things like Python, right, a lot of pros and cons to Python, but pragmatically, one of the big issues with Python is what a pain in the the butt it is to uh to package up a Python sort of uh program with its within with its environment, right? with its dependencies and all the and so there's there's 10 different mutually exclusive approaches people have come up with to this and you know I think what we've kind of standardized on we being the industry is just like slap it in a docker container and ship that um but then you're trying to explain docker containers to the [laughter] to the IT and it's like so you you know it actually nudges us in some cases toward things that are like um oh statically linked compiled executables right that's so like come back to rust or even C++ you know there's cases where it's like, hey, I can leave you this executable and it's going to run for the next 20 freaking years, right? Yeah. And as long as I get it, as long as I get it right the first time, you know, um there's there's some value in that. Yeah. Um you know, we we we kind of build in ways that are a little bit different than what you do if you have a full sort of a uh you know, what's the buzz word? DevOps type team or things like that. Yeah. An existing tech stack that's beyond the email service. you know, we talk about favorite projects. Like I'm a again, this is this has been like the the 20-y year journey of like how do I finally get back into the the the thing I always kind of wanted to do, which is sort of like programming for its own uh its own sake. Um, some of the most interesting projects I think we've had actually are are kind of these weird cross-disciplinary ones where uh we have a very industry specific problem, but there turns out to be sort of this this other lens you can look at it through that looks a lot more like something else that that say uh we've known in computer science since the 60s or 70s. And and like one that comes to mind is how many problems actually turn out to be sort of uh language problems, language translation problems or or maybe better put like how many tools that we need for a lot of different random stuff look sort of like compilers if you squint. Um, and so I' I've worked on some of my favorite projects involve like, hey, translating Aries lines to, you know, uh, production forecasts and and when you get into like the fact that Aries economic lines are are really kind of their own, uh, programming language, you end up building, you know, literally a parser and and something that is kind of like a tiny compiler that uh, you know, translates from from Aries uh, syntax trees to, you know, whatever your output is or um this so I guess I'll pitch our our product. like constantly one year from from coming out which is this thing that grew out of a actually was a I have a friend of mine um is involved in splee the um the reserves guys society of petroleum evaluation engineers right and um they they were interested in a a talk on Monte Carlo methods and sort of so I I gave one back in I think it was February this year on maybe it's been longer no god that was February last year sorry there's been a there's been a whole toddler in between. So that's like brains blackout. Yeah. Um [laughter] but you know and and so the talk I gave was on kind of pitfalls of Monte Carlo and and how you know in especially in in petroleum economics or reservoir engineering some of the things we we often tend to miss like uh you know correlation between uh assets or or outcomes and and things like uh you know modeling the decisions you're going to make in the course of a project. Um, and along the way I I kind of discovered like, "Oh man, I really want to make this like an interactive show and tell. I looked at Crystal Ball. I looked at at risk." I was like, "Okay, this kind of sucks. Like this is ugly. It's slow." Um, and so I guess I I for whatever reason had a lot of time on my hands the weekend before this thing. And I actually ended up building like a kind of a Monte Carlo uh modeling language, probabilistic modeling language prototype with um h kind of like we talked about the you know the the difference between a ripple and like a truly interactive shell. And so it's got this interactive guey where you can you know for every variable in your model you can look at the you know the uh univariate or or biariate stats and and see all these plots on the fly and it's fast and yada yada. So we I've been I've been slowly trying to turn that into a product for uh a couple years now and it just I don't you know in the consulting world everything's very uh time sensitive interrupt driven so you know well in the business is good then catch 22 of like you actually all your product work ends up kind of happening on like bad years so by definition [laughter] yeah um but no that's that's something we're looking at and probably you know I've said this with a caveat I've said this would be industry agnostic that's more like just a money so what we what we really want to do is is build in the um kind of the the the primitives, right? like the distributions and the the operators and the features that would make it very attractive for building uh sort of simple quicklook petroleum economic models and and what we have in mind is kind of like the so it's kind of around what what is hard to model in some of the existing tools and that's things like hey I've got a pilot program and um you know if the pilot exceeds this threshold I'm going to move on to appraisal and if that exceeds the threshold I'm move on to development and hey also like because these are all drilled in the same basin like there's going to be a correlation between my my pilot, right? But the wells within my pilot and wells within my appraisal and it's just it it's I'm not saying it can't be done, but it it's somewhat challenging to model those and and the tools that are out there. And so and and if you don't model them, you you end up um sort of well, it can go either way. You you can end up lpping off a lot of upside because you don't recognize that um actually that's that's the incorrect way to say that. you you can end up failing to lop off a lot of downside because what you've modeled in is like a a program where you're hellbent on drilling. And so if you don't if you don't build in that exit gate after your pilot, you know, you you suddenly your economics look terrible because you've you're going to go full pull steam ahead no matter what, right? It's like the assumption that hey, even though we drill this well or like if the well produces zero barrels of oil, we're going to continue drilling for right keep. never thought about that with the Monte Carlo stuff cuz I mean that's a I worked for uh BASIP for yeah a very short period selling their uh reservoir modeling or reservoir sim stuff and basin modeling stuff and I found that fascinating how many just how many knobs there are to tweak to turn on a reservoir model and like it just overwhelmed the [ __ ] out of me because it's like this is just on a single well right and it's like now we've got hundreds of these and a producing field like you've so many var like there's so many variables on just a well right outside of all the rock [ __ ] which like but then you start throwing those in and we've seen so much over the last man decade of like this evolution of like frackets right and how the the actual operating activity is affecting your production and your new stuff and your old stuff and like it's all one giant system ultimately and so that's that's really cool that you're looking at being able to put in those because it's it's reality Yeah. And it's and you know, you're not going to drill a 100 wells if they all if the first 10 make zero, right? Yeah. And and so and then like cynically, I mean like the the the only problem with any probabilistic modeling exercise in the oil field is we we quantify all these because we're we're reservoir well. I'm a reservoir engineer. I've worked with reservoir engineers and we we have some intuition or some idea for how to quantify the uncertainty like well I think the distribution of procity might be this and I think the distribution of permeability might be this and we're going to build up these hundred variables and simulate the thing out and then it's like okay we're going to slap three different price decks on it right and it's like you know $20 50 so it you the the financial you know the fiscal and and regulatory and I think that uncertainty kind of dominates and there's not a a nice way around that or if there were I' I'd the much wealthier man [laughter] very different place maybe they was it renaissance or some of the quants that think they've really cracked the code um anyway I don't know maybe they've got it but uh but you know we we want to kind of make it easier at least to quantify the uncertainty that we can't quantify no that'd be really cool well just having I mean having a a process or a model that represents more of a realistic yeah scenario right I think that's anybody who has ever used any kind of model whether it's reservoir simulation or any of the stuff that we did in college, right? It's like a model is a model, right? It's only as good as the inputs and it's only a model. [laughter] Like it has to account for all of us, right? Yeah. Correct. Yeah. It's just how right how good at being right sometimes are they came up so much in in data science projects too is like people people fix it on R squar or accuracy and what you really you know accuracy is always kind of bad, right? Like we there's so many unknowns and a directionality subsurface stuff especially. So what Right. Exactly. So what what we were always trying to push people toward is well all right the accuracy is what it is but let's let's use this to build some intuition on what drives what. Right. Like I've got X1 through X10. Let's pull out X3 versus Y and see what that incremental you know or marginal. And then it comes down to also there's the explanability side of it too like where people can actually understand what the hell's going on. A linear regression is way easier for someone to look at thing and understand like right and yeah maybe your you know you know deep learning model might be more accurate but like no one knows what the hell it's doing like you know so if it's within a certain error margin like use a linear model because people can look at that and like understand at least intuitively what the yeah that's I mean don't don't sleep on those simple models either especially for BI stuff right like I've been I worked at a company and I started playing around with some of the Python I think it was or not Python in uh in Spotfire some of the uh regression analysis and stuff that's built in. You know, I was in on the sales side at the time. And so I was like, "Oh, let's throw in like oil prices, rig count, completion, like all the permits. I threw as much stuff in there as I possibly could." Y and guess what? The biggest impact like 80 plus% on our revenue was oil price. And it trailed oil price by a quarter and it was per like it's a great linear. But had I never done that principle and I thought it was going to be this like oh well it's going to be based off rig count or permits or whatever. It's like no it's well that drives all those things. Exactly. There's so many things in our industry where if you if you can kind of transform the variables so they they play nice with linear models or I I also like random forests and and kind of decision tree models because they got some of the same flavor like you can show a right there's flowchart if um yeah I just it it's amazing how we seem to want to jump to like the most complex possible thing and uh you also have to be comfortable just like knowing that hey there are probably other factors in this and I need to be cognizant of them that I might not be taking into account with this. But for what I've got, this is the best representation of it, right? It's never going to be perfect, right? Cool. That was that was a quick hour. Yeah. Pretty fast, [laughter] man. That was very quick. Yeah. This has been really good. Yeah. But we usually uh end with like a kind of a lightning round just Okay. Yeah. Oh, I love it. Try to keep it at one word, but Yep. Sometimes short sentence. Yeah. Yeah. I'll uh I'll jump in one. What is who is one of your favorite or biggest kind of influences either in general or in the tech space or Edgar we can say oil and gas space [snorts] tech Edgar Dystra first Dutch programmer uh was a professor mathemat kind of mathematician turned computer scientist and uh you can look him up he's got a lot of really great kind of grumpy old man memor is he active on any kind of like social years ago Yeah. What's your, you know, favorite uh open source tool or, you know, Vim? Vim. Okay. Yeah. What is that? It's a text editor. It's a '7s text editor. Is that the one that I Well, that I might have found or used. Uh, maybe. I mean, you may have been stuck in it because you couldn't exit it. Yeah, I I I live in No, like you have to I forget what it's like qi or like there colon Q enter or or I think you can capital Z or Yeah, I don't know. But you have to look it up. It's kind of like a meme at this point, like how to how to exit Vim. [laughter]

Um, what is one of your favorite books? Both in general and techreated or engineering related, either one. Okay. Moby Dick. Um, everyone should read Moby Dick, but um, tech. Oh, that's a that's a actually a challenging one. Let me It's like I'm I'm trying to You give a couple design. I'm gonna I'm gonna can't narrow it down. It's hard.

Yeah. I don't know. I'm going to punt on that one. I know it's terrible, but I'm like I'm kind of drawn to a blank. I'd say there's there there are reference books I come come back to over and over again, but a lot of them are like types and programming languages. Ben, but that's a if you go on my GitHub, you'll see I have weird hobbies, so some of these answers aren't going to necessarily make sense in our industry. No, that's why I love that's why I like it. Yeah. What's your uh what's your favorite like managed service? Oh, so you're you're heavy like coder type. So I I try to avoid them. I I run our So I I Lenode I've had good luck with Lenode over the years um for for our servers. They just got bought out I think by Akami. We'll see how that goes. Okay. Um AWS confuses and frightens me. [laughter] I I love Terraform for one reason which is I don't have to know AWS CLI. I can just kind of like have a But I also try to avoid being in a situation where I need that. Right. That's true, too. What's uh one thing you could nerd about nerd out about and spend hours talking about? Oh, uh [sighs] I mean a bunch of things besides what? Yeah, besides tech. Yeah, maybe something outside of tech engineering. Oh, probably uh like military aviation. Okay, cool. Scotch Scotch distilleries. Scotch distilleries on in the islands off the coast of Scotland. Nice. Um, that's perfect. I'm I'm right there with you on the whiskey side. But, [snorts] um, make a prediction of what the future of energy or energy tech looks like in 10 years. What's one of the biggest impacts you think that's coming down the pipe in the next 10 years? [clears throat]

In energy specifically, man, I hate making predictions about the future. The beauty of it is that no one's no one's here data science, right? You should So I what I what what I maybe this is more of a hopeful prediction. What is your linear regression model say? I think what needs to happen is we need to pull data science, data analytics, data management even to an extent back out of being seen as kind of an IT function and and pull it back into being a core engineering or or business or or basically see it as a profit not not only see as a profit center rather than a cost center but also the people managing that and frankly the people doing that work dayto-day need to be the domain experts not kind of um so I I It it ties into longer term like you know predictions about well is the tech industry kind of in the middle of a as as we are going to learn what's Warren Buffett saying like when the when the tide goes out you're going to learn who's been swimming with no pants on right so like as we kind of come out of a zero interest rate environment and venture capital gets tighter. I think you're seeing contraction in the tech industry and I I think that's going to drive some people to look elsewhere for employment. I think we may pick up some talent from that which I think can be a good thing. But you have to upskill them on the But you've also got to bring them in as domain experts. Cuz the last thing we need is a bunch of ex Googlers coming in and saying, "Oh, I know what y'all need." Yeah. A freaking Kubernetes cluster [laughter] when the real problem is like, man, I've got, you know, LOS statements over here and um, you know, [snorts] basically, you know, you're you're trying to do what else do a VLOOKUP and some guy comes in and says like, oh, let's throw Kubernetes at it, right? Let's not [laughter] just yet. But I think what you know maybe hopefully what I what I want to see is a little bit of a rationalization around kind of you know we've been so we've gone from being kind of a slow adopter to frankly I think an an overaggressive adopter of like the shiny new thing and and how many companies have jumped on whatever Hadoop that's a great one like I dodged the freaking Hadoop I my my career like and maybe this is just technology in general because I haven't been in you know I'm not that old but um I mean I birthed to death of Hadoop man just Like I remember what they were pushing the [ __ ] out of a how fast it died was amazing too cuz it was like I mean I was you know it was 2014 and I was starting to dabble and like trying to figure out what technology and I kind of pushed that one off and then like all of a sudden it was like two years later like no one talked about so you know it's kind of a broader thing but like I would love to see us people I I would love to see people focus more on the fundamentals like understanding relational data and understanding you know what are the advantages of of row oriented versus column oriented storage good quality data. Yeah, [laughter] that's that's eternal, right? But knowing the details of Hadoop or Snowflake or whatever next year's thing is like I mean yeah they very much so it matters but it's also kind of a femoral right and so we they get forcefed you know the new shiny things by the consultants and the big three and all those guys and it's it's good to be cognizant of it but don't go tell your devs we need to implement this today because Gardner said it. So I hope we kind of calm down. I hope I hope we kind of get some maturity and and say, "Well, all right, we don't have to chase a new shiny, but let's let's think about what we're really trying to do and and what makes sense for sure." So, on that kind of what's your what's a bleeding edge tech or something that you're actually pretty bullish on? Okay. Um, chat GPT has been like 80% of this answer. [laughter] So, you Yeah. So, all right. I know you have opinions on I'll give you a hot take on AI. I I don't think Chad GPT is going to take programming jobs anytime soon. Um, I so my wife and I actually did a we we did a terminus webinar back over I think over Christmas break on Chad GBT 3 when that first came out. And you know we we do this a code is this kind of yearly coding competition that we we kind of do the first like week of every year and then get you know takes hours and hours you like forget this. Um but we actually we started doing this competition like John Henry versus the steam engine where you you know she would run GPT and then I would try to solve the problem. first three days kind of they're like basic interview level questions. It it actually smoked the human. I mean like the the leaderboard was all um you know GPT generated code. By like day five or six though it was it was failing. And and what what I've noticed is it it really just it is really good at producing an answer that looks superficially plausible and you like zoom in and look and you're like this is nonsense or you copy and paste it. I don't you're like, you know, I I see it as kind of like a I use it almost more like a search engine because Google Google's gone to hell chasing the the SEO thing. So, you know, I'll I'll type in like, hey, I'm looking for a plotting library for Rust that can show me, you know, heat maps of uh spatial data or whatever and it'll say like, oh, it's called blah blah blah. And then half the time that doesn't exist, but you know, whatever. Still better than Google. I think unfortunately that's the scary part of GPT is it is sometimes. Yeah. As far as what I think is promising, um, dumb programming language nerd answer is like I I I love that we're living through the resurgence of static types and and functional programming. I think Rust and even like C newer versions to an extent are taking those ideas and making them mainstream. And I think it's a nice correction in this eternal pendulum of like Java to Python to um, you know, hopefully [clears throat] something that looks a lot more like that. That's that's going to affect the way we do things a little bit. I I think Python's not a not a a sustainable substrate to build uh engineering tools on. I I think it's comical. It's kind of become the language of engineering. You know, when you think about what do you need? Well, I need correctness. I need fast uh you know, I need good performance for floating point code and I need um you know, I need something I can leave on a server and run for 20 years. Like, well, Python checks none of those freaking boxes. [laughter] Um, so I would love to see, you know, you know, like I said, kind of a return to, hey, let's let the computer help us with code correctness. Let's use static types. Let's, um, you know, let's let's take advantage of some patterns that minimize the the complexity of the systems we build. Uh, I know the answer you're fishing for, so I'll say because I I agree this is this is interesting, too. It is interesting to see kind of a a move. Um, you know, we live through the wave of NoSQL and kind of the document store thing, which boiled down to and I I had this saying I probably said over and over and I've had to say it less recently, which is like there's no such thing as unstructured data. There's just data you haven't sat down and figured out the structure yet. Like it's it's kind of like um Well, yeah, [ __ ] is an easy button why figure things out, but eventually you land on a scheme. It's the equivalent of of as a kid, you know, your mom says clean your room and you just like put all your [ __ ] in the closet. It's like, yeah, it's clean, I guess, but then at some point someone's going to have to go in there. Yeah. Yeah, but your shirt's up and it all falls out, right? So, you know, I'll say no no SQL circa 2010s didn't impress me that much. I do think it's interesting to see though some of the newer generation of like column oriented stores. And so, I I would in the same spirit as like the weird, you know, uh skitso board full of pens connecting James C. Scott and and Lisp [laughter] and I I see the same thing when I think about like what's happening in game development and and a move away from uh kind of these big objectoriented uh simulations to things like uh entity component systems. Um, same idea. We're we're we're seeing across a whole category of of or multiple categories of application that sometimes it makes sense not to look at things in kind of a a row oriented or object-oriented or um you know here's this is a well and it has these five attributes and this well has these five attributes but instead flip it around and say here's these five attributes sorry and uh and these these 10,000 wells all have those attributes because that that plays a lot nicer with modern computer architecture in terms of performance for for doing certain operations right so when when we talk about OLT P versus OLAP. That's fundamentally one of the big distinctions there is like do you need to do do you need fast uh updates and inserts and and reads on a kind of a row level like show me uh here's 10 here's 10,000 transactions at you know this Walmart time series or do you do you need a fast operation of um total up you know um you know total category A sales plus total category B sales plus total category C sales across all Walmarts in the world. Um, and and and there's a trade-off there. There's, you know, in terms of how you orient the data. And so I do I see things like duct DB and uh other cloud offerings which I won't be clown myself by pretending to know which ones are oriented which way. Um, yeah, I mean most of like the the big ones you like Snowflake, BigQuery, Red Shift, all the modern BigQuery is a columner from what I understand. Yeah. And so that's I think that makes sense for certain applications that we have. Well, awesome. Well, now I'm in where uh John's got No, I was just gonna say where can people find you? What's your Oh, uh pretty easy to find. If you let you know our Terminus website, I've got a blog at usethe.computer. It it doesn't often have to do with oil and gas, but a lot of interesting programming tidbits where we try to uh mostly I try to take kind of interesting things from u the wider world of uh of programming or computer science and and make them semi-accessible or at least uh in Python, which is kind of what we have to do, I guess, to reach people these days. Um but yeah, so LinkedIn um GitHub, Derek Turk on on GitHub. I've got a bunch of of things out there uh in various states of completion or not. And some of them are even oil and gas related or or energy related. Yeah. No, that's that's uh there's you got quite a bit of stuff on there. It's good. That's why I want that's why I brought it up so people can go check it out. Uh what's what's the website uh for Terminus? Oh, Terminus terminusdatascience.com. Got it. Yeah. Keep it simple. Yeah. Appreciate it so much. Yeah. Yeah. Thanks for being on. And um also, you know, make sure if people like the uh yeah, if you guys like what we're doing, please go uh give us a review, like and follow the uh the DW pages, and uh we'll see you all next time. Appreciate All right.