Jake Mannix is currently the Data Architect of Search Relevance at Salesforce. In other words, he's an AI Engineer. Jake lives in the intersection of search, recommender-systems, and applied machine learning, with an eye for horizontal scalability and distributed systems. Prior to his current position, Jake was the Lead Data Engineer in the Office of the CTO at Lucidworks.
In college, Jake studied algebraic topology and particle cosmology. He's ABD ("all but dissertation") from the University of Washington before transitioning into technology.
Some of Jakes previous accomplishments
Jake Mannix is currently the Data Architect of Search Relevance at Salesforce. In other words, he's an AI Engineer. Jake lives in the intersection of search, recommender-systems, and applied machine learning, with an eye for horizontal scalability and distributed systems. Prior to his current position, Jake was the Lead Data Engineer in the Office of the CTO at Lucidworks.
In college, Jake studied algebraic topology and particle cosmology. He's ABD ("all but dissertation") from the University of Washington before transitioning into technology.
Some of Jakes previous accomplishments
Grant Ingersoll: 0:19
this episode Develop Mentor is sponsored by This Dot Labs. This Dot Labs is a framework agnostic consultancy firm that specializes in JavaScript and offer service is including open source consulting, mentorship and training for teams and individuals. If you would like to learn more about their service is visit their website, this dot labs dot com. Welcome everyone to the development or podcast your source for all things, careers and technology. I'm your host Granting your soul For those of you new to the show, we have two simple goals here. One. We want to highlight interesting people in technology across a variety of roles, including, but not only engineering. And two. We really want to showcase all the different paths people might take to get into those roles. So whether you are a new to tech or looking to change careers within tech, we hope this show might just help you find your path. In fact, today's guest is a great example of someone taking a slightly different path and engineering. He's a self professed math nerd who is, as they say in academia, a B D are all but dissertation in physics. Our guest has worked for a number of large and small tech companies over the years and has been primarily focused most of his career on this ah, niche space that I also happen to work on, which is search and artificial intelligence. Please. Welcome to the show, Jake Mannix. Jake. Great to have you here.
Jake Mannix: 1:55
Thanks for having me, Grant.
Grant Ingersoll: 1:57
You know, I think you know, if you spend any amount of time in tech, you know, at some point in time, you're bound to meet a physicist turned software engineer. And it's almost always the case that they work on really hard tech problems. And you know, Jake, I've known you for a long time, and I'm pretty sure that glove fits pretty well. So why don't we just have you start off by filling in the audience on you know, your your version of that story?
Jake Mannix: 2:22
Yeah, sure. Um, yes. So kind of the long, the long the short of it is, I kind of danced in and out of physics on math, basically, ever since I fell in love with it and wanna decided I wanted to be a physicist. My grave grew up at around age seven, so I I joke that I I wanted to be a physicist since before I could spell it. Um, because I'm not necessarily that fantastic, speller, That maybe doesn't say very much. But somewhere around age seven, um, and it wasn't until trying my hand at grad school in physics and actually been dropping out and then going back to grad school for math and then dropping out and getting a consolation master's degree and then going back into grad school for physics that I did about four in orations of that before I finally said, You know, if I'm if I'm gonna be able to actually, you know, settle down and have a family live in one place and actually develop a career with some stability, I I'm gonna need to actually try something where the job market is not 100 people applying for one job. Um, and as I've been doing a lot of kind of just I've never really taken, You know, I also, you know, joke that my computer science background is such that I took a and still, to this day, I have taken one computer science class. It was intruder programming and see our Capstone project was, uh, you know, Conway's game of life. If I ask e printouts. Um and, uh, and that was that was in, Well, let's just say it was in 19 Something something I just won't say exactly when, um t to help my career ST stay afloat, as I say. Um, but, yes, I took Yeah, Unterberg Amy and seeing. Other than that, I never I was math and physics. Some math and businesses, new math and physics. People are hard core programmers, really, at a core as undergrads and grads tubes because they're doing you know, I worked you work with experimentalist former experimentalist who was a oh, stock a cz part of, uh, you know, particle accelerator experiments for a while before she switched into into software and data science. And people like that have been doing, like, deep, low level C an embedded, you know, software for a long time. And they're pretty hard core. Um, I, on the other hand, was a white board guy. I was the kind of person who prided myself on having my undergraduate. You know, senior thesis in physics was 100 pages, or Maur and not a single decimal point. in that. And that's that. If you think about it, decimal points are kind of like numbers exist, but for me, everything had to be a variable. That was some constant. It's a constant of nature. I don't care what the number is. That's not important. Um,
Grant Ingersoll: 5:23
significant digits are overrated, for sure.
Jake Mannix: 5:26
Exactly. You don't Don't don't even knows. It's just it's just alfa
Grant Ingersoll: 5:30
eso How then? You know, you know, and well, especially these days, like a modern mathematician. Modern physicist. I mean, these these folks, right? A lot of code, right? Because it's all it's all simulations and and things like that. But, you know, you take this one class and and I think, you know, looking on your bike you did some work, and I think you just mentioned you did some work at Ah, particle accelerator. Ah, how then did you transition into coding for a living?
Jake Mannix: 6:01
Actually, at first it was I got a job as, ah, software tester. I was hired to work at Ah, real networks back in the dot com boom win Dato when they would hire, you know, anyone with a pulse that you would be a warm body in the seat. And if you actually knew the limits command line because, you know, everything undergrad computer labs were all, you know, spark workstations. And and that was how you got to use computers at school back in those days on. So I was handy with, you know, bash and so forth. And I was like, You could do bash, and you have. And you have a ball school. You're hired. And I actually was just testing, you know, originally. Ugh. Why? Testing And then, um uh, server testing. And then they found out that actually knew how to program and c and new enough math to knew how to be dangerous with encryption. And so someone said, Yeah. Hey, why don't you help write us? Ah, encryption layer for, um, for our video streaming software. I work for real networks in Seattle for a while, and ah, I a kind of transition, kind of sort of by accident into doing a little bit of that, then was laid off because I found the extra didn't need anyone doing that. Um, and they laid off about 30 from the company. It was 2001 hard time s.
Grant Ingersoll: 7:27
So then coming out of real networks. You know, like, you know, as I mentioned, you've got into search and a I And of course, both of those air, very math intensive fields. But you know, what was the break into that field?
Jake Mannix: 7:41
Yes, that actually. So this is actually where when I started learning, I actually remember when I first learned about this whole T f i d f phrase. I was in a little company where they were doing. They were they were actually doing search over podcasts, and they were doing search inside of the text. This is back. And you know that, you know, early two thousands time frame off, searching inside of audio was not an easy thing. Um, and even nowadays is pretty cool when you can do it well, but this company was founded, and I was not working on the search side of that. I was not a search guy at that time, but I remember reading through the code and seeing how you know they were trying to cluster together these thes documents of the taxed and unattended out of the speech to text Andi kind of grooving them, clustering them by the FBI the afternoon like this I d f thing seems really add hawk and weird. That doesn't make any sense at all. What is it for? And it's funny looking back on it now, Thinking of it is like, you know, the bread and butter work horse of our industry. But I learned about it, and and I learned that there were other approaches doing clustering of vectors that were very similar to what I done in in the math world. But, um, you know, matrix math wish I've always been very fond of was the way to do that. I'm like this. This is This is great. This is a solicitor computer. Can you this way faster than I would do by hand. We can write code to do this. Um, the people I was working with, on the other hand, Well, like a major season, could we avoid interestings? Can you just do it some other way? Like, really, this will make it easier. And that was that was that was the sort of thing it was. It was the math and applying it to problems.
Grant Ingersoll: 9:19
And then so, you know, on the show, obviously we're trying to highlight. Ah, lot of different roles into this and and, of course, the T f I d f. Don't worry, listeners all link up in the show. Notes. Ah, some some links to what all of that stuff means. But, you know, like a CZ you think about the programming and day to day work that you do. I mean, how much of it is actually like hard core math versus what you would just call it? Just you know what we might just call just good engineering skills or good software programming skills.
Jake Mannix: 9:54
Well, um, I guess it depends If you wanna measure relative to whom is so thrilled people who are supposedly supposed to be doing applied machine learning work and data science relative to that group of people, I'd probably say, you know, maybe, you know, out of the media, I may be in the 25th percentile of a percentage of my time spent on on hard core math stuff. So not way down in the tiny set, but like, less than average. But then again, we may all think that because we all think that, you know, I'm doing, you know, telling my week probably I don't know, uh, 10% or less is spent doing anything. Really? Matthew related, even though really ostensibly. That's kind of you know, the main reason why I was hired.
Grant Ingersoll: 10:45
Yeah, it's still it's still pretty good chunk of that. And by the way, I just love the fact that all of your answer they're all had math formulas like will, so
Jake Mannix: 10:57
all the time. But that's my point, Asai
Grant Ingersoll: 11:00
said. You know, self professed math nerd. And you know, Hey, I'm I'm one too. Uh, yeah. Okay. So I mean, like underpinning all of this kind of stuff. Obviously, there's this need tohave an understanding of the math, but it's not necessarily a blocker. And and fair to say that, you know, like it's something you you feel comfortable. You felt comfortable learning as you as you went along, right? I mean, you were able to rush up and understand what you needed to do in orderto figure out the
Jake Mannix: 11:33
core principles that not so for me what it was really was. I had to learn all of the software engineering, good practices and the soft development life cycle, and, you know, programming skills and learning, you know, people that take a siesta grease undergrads. You know, they do at least one course on real functional program in ml or 11 of these, like you serious, you know, small talker something even if they never use it. They've been exposed heavily to a variety of different kinds of environments for for for programming languages as well. It's kind of programming environments on guy. Never did. I had to pick that stuff up as I went along. If the parts that I was able to leverage from my previous life was, I'm the math is never the scary part for me. That's the part that says Okay, well, maybe that's gonna be hard to implement, but yes, weaken, we'll figure that out. That's not a problem. Um,
Grant Ingersoll: 12:35
but it's it's amazing. You know how much the math of physics and, ah, you know, engineering like, you know, things like mechanical engineering, physics, et cetera. It's pretty much the same math for search, an aye aye and all that kind of stuff. Obviously, you know, there's a lot of hand waving going on there, but you know, it is that kind of the case done of why, for instance, we see so many physicists who you know, feel right at home, and the software engineering world is is, you know, they're so used to just dealing with these kinds of this kind of math and this kind of systems thinking
Jake Mannix: 13:12
and all of that. Is that you? I'd say the latter. I'd say what you were saying there is. It's the systems thinking it's the model building the notion of this abstraction layer, you know that that is the heart of both. Being a mathematician and physicist is, and in fact, many scientists. I don't know about all scientists, but certainly once you get toward the end of the spectrum for its math and physics, it's all about take a big, messy, complex system, boil it down to two variables or three variables that interact in a way that that replicates the key aspect of the complexity. And that's the whole notion of building a data model building, you know, an abstraction layer writing a system that has decoupled components that interact in a very simple way that you can predict this exactly deterministic state, you know, you immutable, immutable variables, you know, being able to handle things that are conceptually simple on a building block level, and it's only when they become a system. Do things get much more complicated. That's reductionism like that. That that the whole thing of the past four years of signs
Grant Ingersoll: 14:26
Yes, sir, people coming into this kind of role, you know, they need to have this, you know, this, this mix of liking to think, conceptual levels and abstractions. And then they also need this very practical, like you said, of having to learn and understand. Like, how do you actually build working software that can often be quite messy? You know this theory and practice being two different things, but it's stuck a little bit about, you know, looking at your your LinkedIn profile you These days you have this title of architect of search relevance. Ah, and you know, that's that's admittedly a pretty niche role in a company. I mean, what is the day to day of ah, somebody focused on search relevance at a company. Ah, look like what? What do you think about? You know, as you you dig into that
Jake Mannix: 15:18
role. Yes. So it's you know, I I you know, I've only been t give tiny bit of background of Bennett and Salesforce, a search all of its architect for only past three months now. So was my real role. What does it really entail? Is something that, you know I will. I will be myself learning about a ZAY help kind to find the Rolande grow into him more. But I've been doing search, relevance and search infrastructure on and off for the better part of the past decade. And so this time I'm in a large enough company Salesforce having 30,000 employees or so, which is two orders of magnitude larger than the biggest company had ever been at. When I first started out, some of them got up to, you know, 10% that size of the time I left, but but 1% that size when I started. So having someone whose job it is to be the search relevance architect, the person who is supposed to figure out how we do search relevance and you're not doing search relevance infrastructure. You've seen someone else's going machine learning model pipelines. Someone else is doing machine learning. Uh, you know, search, relevance and search Ranking model serving, uh, model serving infrastructure. You've got another person doing. You know, a lot of the batch compute another person handling all the search indexing and all these things. All these systems are already handled by other teams, and they're just search relevance. Is this specific goal of, you know, making sure that what we're showing to users when you enter your search queries to do a search engine are the relevant ones. Um, what does it mean at a level of being architect? Interestingly, I thought it was much more of, ah, individual contributor role, because definitely I don't have any direct reports in this particular role. I've been a manager in the past, but and this really don't have any reports that Okay, great. I'm gonna be coding a lot, actually. As it turns out, when you're in an architect type position, high level, your job is to know a bunch of stuff and help make sure that other people don't waste their time doing things that they don't realize. There's actually something that there's a way to do it easier. Um and so you spend a lot of time in the same meetings that, you know, you think that the managers, the ones all alone meetings, but actually the architects of unity and all these meetings to to help make sure that people get directed to the right resources and they get linked up together. And sometimes, you know, I saw a funny tweet the other day of someone in a similar role to mind saying No, I was really proud of, you know, felt really productive This Monday I responded to a long email thread and ended it by giving someone a link to some documentation that they really needed to see. And that was the most productive thing they did that day on. They weren't being ironic that they weren't actually being. That was my war like, this is how useless I have. They were feeling proud that they actually solve this whole email threat of many people who are confused by a complex topic. Found the documentation of this person said Yes, this answers your question. We're done here and it's all the problems.
Grant Ingersoll: 18:22
Hey, work, work smarter, not harder. Well and then so then the relevant side, I imagine, though you know a lot of it. You're also looking at the day, you know, try and understand the data And in fact, perhaps that ties this into, ah, your time at lucid works, which is where we last worked together and and some of your prior roles where you've had this title of data engineer. Yeah, And I think this this title is is an increasingly influential role, as people look to get machine learning and things like that and into production. And, you know, perhaps, you know, reflect a little bit on what that rules all about and perhaps comparing contrasted to roles that others may have heard of, like data scientists or software engineer like, how
Jake Mannix: 19:09
do you like being a data engineer? Yeah, So if if the if so tend to kind of give some context if if people say if you have this this this coming, you know, you know, diagram that says O R I guess it's the quote from Josh Wills. Uh uh, That Ah, um uh, the you know, the data scientists is someone who's a better programmer than any statistician and a better statistician than any programmer. Um, uh or or if you want to be self deprecating, it's It's the person who is a worse, uh, statistician that any minute he's verses. Is that statistics? And a statistician and worst of programming that any program. But either way, halfway in between a data engineer, I would say, is someone who's halfway between a data scientist and a software engineer. Um, so you do need to keep your hand on the mission learning side of things. You do, you need to know what are the inputs to, um, you know, Ah, clustering algorithm, or how is it going to scale? Um, but a lot of the time, you're also the one making sure that detail pipelines to get data into a system are sensibly set up. And and you know whether or not you're supposed to be setting up a streaming operation or a batch operation. Um, date engineers need to think about both the data warehouse and the kind of the data at rest, as well as how it lives at a run time in the system. Data scientists usually are pretty uncomfortable with working in runtime systems, a place you know, if I'm gonna qualify runtime systems being a place where you might get paged because a website went down date engineers need to be comfortable with that. They may not be directly on that front line, but in general they may be the person helping to make sure that whatever new instrumentation happens in the U. Y is getting the events to get sent back. And when someone clicks on a page, you want to make sure that event is recorded with the right context so that eventually later someone could, you know, either track the metric or train a model. And if you're not tracking the events, you're not instigating it, right? You're not gonna you know that, you know, be able to do anything you want. You'll be flying black.
Grant Ingersoll: 21:25
No. Yeah, that makes a lot of sense. I mean, I think it's like I said, it's such an increasingly important role in And you know what? The sheer scale of data that people often are dealing with these days Having somebody who who can kind of straddle both of those lines makes makes a lot of sense. You shifting gears a little bit. One of things I like to do on this show is really kind of OK, This is how you got to where you are. Now let's talk a little bit about where it's all going. Ah, you know, I know it's obviously hard to predict the future, but, you know, you know, as you look at this the space of a I and search and recommendations and kind of all these areas that you've you've been in Ah,
Jake Mannix: 22:10
what do you
Grant Ingersoll: 22:11
see is kind of the big opportunities and challenges, you know, for for people like you, in terms of being successful in the
Jake Mannix: 22:20
role, Well, I think I think that some of the upper the biggest opportunities are in the fact that we've come this whole notion of a I being Maur democratized. And if you've got data, um, you don't need to be a, uh you know, an ai AI expert to train systems these days. You know, if you look at some of the open source, um, packages and projects that come out of, you know, the one of my other former employees, the Allen Institute for Artificial Intelligence, you've got these. You know, this Elmo project, which builds contextual word vectors. They're kind of building up semantic meaning from text. These packages is pre trained models. Are you know, not just downloadable, but you can kind of import them into your project with, like a one line import statement that says, I want to get Elmo Version one, large one and you imported into your pipe on project and poof. You've got the ability to have contextual word vectors imported into your machine learning pipeline. You didn't need to try anything. It's already pre trained. And now you can use that to augment what you want to play with and similar to the ones you know from from From from Google and Facebook Has this one's come out as well? And these kinds of models both exist in a service later. Where do you say? Okay, I wantto hit, Um uh, 10 point on in my Web and and I'm already I have a cloud Web app and I want to train up a model where I want to use a model that has voiced attack so I could just call a Navy I hear on Amazon are on on G. C P or or using 1/3 body surface like Al Gore is me. Um, the the flexibility of being able to do this kind of stuff without being a machine learning expert means that if you're willing to not be afraid of it and just kind of go in and see if you can just tinker with it. Treat it like a software system the same way that, you know, 20 years ago, you were like, Oh, wow. Hey, look, there's already a software library to do this, you know, token ization of text thing. This this Lucy in library exists. I could just Wow, I can even token eyes. Chinese. I don't know anything about Chinese, but there's already a token Isar for it. Um, and then you could use it as a library. Now we can do that with, you know, contextual word vectors, speech to text. Captioning of images automatically. That kind of thing means that you can use them as building blocks and build something even Maur fantastic. Out of those, um and so I see that the possibilities for the future as being one in which you treat, you know, a i libraries, as if they were things that you should be able to kind of inject in to your system. Um, kind of wherever you think you you can make use of it. And so getting staying abreast of what is already available out there now and really kind of works is kind of a lot of work because it's things they're coming out every month. New stuff. But really, the possibilities are pretty fantastic without having to be even. You know what a cutting edge, you know, deep learning researcher yourself, you don't need to Their already people open sourcing these libraries, pre trained models being ready to be just used.
Grant Ingersoll: 25:34
Yeah, and the beauty of them right is the quality is often good enough, right? Like sure, if your Google and you need to make you know that extra little decimal point on the end. Ah, you know, you really spend a lot of extra time on it. But for kind of the rest of us, you know, a lot of this stuff could just be incorporated in. And then I imagine then that you know, one of the biggest challenges you see it in your role Is this like, you know, this is just how do I keep up all right, of you know, Hey, there's a new you know, the flavor of the week in the A I space, right? And then, you know, imagine too You know, there's a lot of work that you need to do in order to determine Winner These things behaving well, when have they gone, arise? Maybe spend spend a couple of minutes talking about kind of the how you think about the challenge of ethics and a I and and bias and things like that.
Jake Mannix: 26:34
Oh, yeah, that's a That's a real tricky one. I
Grant Ingersoll: 26:36
know that we could go on for a long time on that one. So maybe you could sum up some of your key thoughts there.
Jake Mannix: 26:43
Yeah. So, um who, um that's so I've Yeah. When you especially have public facing features, that's super important. Um, how do you I don't know that that I think that's something that I don't have a really great answer to. Um, you know, because they can creep up all of you. I could list in a series of anecdotes of times when it's happened where the fact that you trained on English language text and you don't realize that that the number of times you get words in Spanish are rare, and then suddenly it's gonna it's gonna have a particular location for what it sees. Spanish text I'm just do the device and your training and smelly bias bias. But it's just Ah, yeah, training data. You know, the frequencies of generations of these two in your data set and data set cleaning, which is a huge job for Dean. Engineers really is spending a lot of time. You're cleaning up data. You let that stop us sneak in. And this the strange bias you didn't think you have, um could be there. So, um, you know, at least you know, different companies. I think one of the biggest things is it's important, actually. Have someone whose job it is specifically to think about this. Obviously, we all need to think about it better. But having someone to go to ask and say, What do I do about worrying about training data for this kind of classifier and having one of your company who actually thinks about that at a at a serious level? And in big companies, you can actually have, like, you know, an office of, of humane, you know, research or whatever, But that thing's company exactly an ethics board.
Grant Ingersoll: 28:21
We're gonna have the Maybe we're making up a new name here of the ethics engineer.
Jake Mannix: 28:27
It's true. And ideally, data scientists will be thinking about this already. But this stuff is is hard to know, you know, way run to that.
Grant Ingersoll: 28:39
Yeah. Now that I think you hit the nail on the head there. And so you know the number one rule right is just be conscious of it, right? And then and and know that you even have your own bias built in you, no matter how you how much you think you are, aren't there. There's some level bias somewhere, right? And it's not always big, you know, evil, societal biases. It could be like you said, things that just creep in from from, you know, small things like in search. One of the easiest runs is like people will click on the first thing before the click on the second thing, you know, all else being equal and so you can make judgments based off of that that, you know, screw up everything else
Jake Mannix: 29:20
downstream on more insidious. You have things that if if there's ah you know people's eyes, they think they know your eyes had such a huge effect on what you do. You see a thumbnail next to the image next to the results. And they're so heavily, you know, just trained to see something that's either provocative or strange or maybe even bad in the picture means they click on that result and and then you suddenly, you know, you can optimize for the fact that now you start saying you start showing, you know, MME. Or inappropriate pictures just because those actually get more clicks and and and some of your systems like, I don't know what the signal is that says these should go to the top. But, you know, they we keep getting the click. So we're gonna keep sending to the top and, uh, you know, that's not so great.
Grant Ingersoll: 30:05
Eso One of the things, baby, for people going into this really is just to think about, you know, how do they make sure they check their their ethics and their bias cities, as as they build this out? So, you know Hey, Jake, you know, this has been awesome. I love this story here of physicist turned ah, turned engineer. Lots of great great little insights and tidbits there into your career. And and, you know, I think in the interest of my goal here. I want to I want to try to keep these podcasts around 2030 minutes. And so, you know, first off, I just want to thank you for coming on the show, but then I wanna finish up with my my one final question that I asked everybody on the show. And that's, you know, you kind of take that step back as you as you reflect on your career. What advice would you give to somebody who wants to get into this space of search and a I and data engineering and and perhaps architecture? I
Jake Mannix: 31:06
would say, um is for people that have an inclination in this direction because they happen to be pretty mathematically inclined. Um, I would say the focus you should really do is becoming a good you know, software engineer. First, go into it from the perspective of having good development practices in a sense of, you know, organized projects and learning dependency management properly and actually suffer engineering stuff. You don't learn in school in a software Internet suffer, you know, you know your science program, but the stuff you learn your 1st 2 or three years as a straight up software introduce. Forget it. No. All the fancy machine learning and a I stuff. You get that background coupled with a strong mathematical foundation of abstraction, Um, and then is the step forward into becoming a date engineer or relevance architect or or machine learning engineer is not hard at all. That jump is actually straightforward, so I can't speak to the people that want to come into it from a. I've learned stuff from a technical standpoint from software engineering, and I wantto learn more of this math. There's a whole separate kind of path, and people will give you a conversation about that and because I can't relate to that coming from it from the math 1st 1 of you, I say, The very first thing you do is don't try to say, Oh, I gotta shore up my better stats. Background. Frankly, I've never taken a physics class in my life, even though I have a master's in math. Um, it's sad, but it's true, even though that's the most Matthews now, his statistics, other than winning a relative and that's that's what I say is is don't worry about that. If you've got your background is mathematical started, and you want to get into this engineering focus on the fact that you can always pick up math. You know, no problem later on, even if you don't learn that particular had a math, because you know how to do that. What you don't yet know how to do is being the software engineer, and that's important even for data scientist. Whenever I try to hire data scientists, I make sure that I can tell that they're probably pretty good, well organized software engineers that happen to be doing data science. If that's if that's what they are, then they're they're hired in my book. Um,
Grant Ingersoll: 33:22
yeah, I just, you know, a lot of really good advice there, although it's funny, I I have a math degree and I will give one little ah tweak on there if if If you don't mind. And that is just make sure you still stay up on your math as you go is well because, ah, it's pretty easy to get rusty on some of those things to prove it. It is true a zay can attest. I Sometimes I go back. I like Wait, I didn't do that math. It always takes a little bit. The unwind. I'm like, Yep. I still know how to do that. Yeah. Ah, well, Hey, Jake. Very awesome. Toa have you on the show again and catch up. Ah, just appreciate so much. You joining the podcast and and best of luck in this new
Jake Mannix: 34:10
role for you. Thanks so much. Thanks, Grant.