0:00 Welcome back everyone to another, another glorious episode of energy bites, John Kalfian here, my trusty, uh, co host, Bobby Neeland. What's up? What's up? Collin in from Southwest Houston.
0:16 I don't know if I'll be there later, but I just couldn't, couldn't make the trip as early yet. I don't, but it ended up being easier since Michael's remote to, we've got our remote set up in the
0:25 studio So
0:28 no worries, Bobby and I are going to geek out over some database stuff later this afternoon. So excited to, to get to brain slap each other with that. But today we've got Michael Egan, he's the
0:40 co-founder of Bunting Labs on the show today. Michael, thanks for, thanks for joining us, man. Yeah, thanks for having super excited to talk more about maps and energy and how they work together.
0:51 All things maps. I love it Yeah, I mean, I don't know how much you know about the oil and gas industry, but we've got people who love maps. I mean, especially like geologist and I mean, the
1:02 engineers, the plan man, it doesn't end. So if you can integrate with GIS, it's a big deal. Yeah. I sent your site to our new customer service manager. He's an exgeologist and I sent it over to
1:19 him. He's like, I've got a long list of questions. I was like, Don't worry, don't worry, we'll get those questions asked for you. But Michael, why don't you just start off telling us a little
1:29 bit about yourself I don't know where you're from, where'd you go to school, how'd you ultimately end up at bunting? Yeah. So, I'm from LA and went to USC, studied in history, which I guess
1:45 we'll get into it. It doesn't really fit with a mapping AI company, but here we are. But yeah, so I was at USC and one of my longtime friends have actually known him since elementary He went to
1:60 MIT for college. And during the pandemic, we're all sitting at home. He texts me, I have a billion dollar idea. Do you want to work on a startup on it? And as a spoiler alert, it was not a
2:12 billion dollar idea. But I got us into this, which was fun. And we started working on AI for predicting rent. And so I don't know if you've ever seen the company co-star, but they aggregate and
2:24 sell commercial rent data. And we were making AI to try to predict what rent will be at a building and sell that to real estate developers. And it turns out they don't want the AI data. But it kind
2:38 of got us into thinking about spatial data and thinking about sort of the power of it and how it is not really well utilized. And so we started moving along with different ideas. And that's when we
2:53 started speaking with sort of traditional GIS analysts. USC has, I've. learned after graduating actually a very big GIS program. It's one of the - I think it's one of the only schools that offers
3:04 a standalone master's degree in GIS. And so there's a big alumni network of GIS analysts that I could go speak with. And they kept bringing up different problems that we thought we could solve with
3:16 AI. So we, since about 2022, have been working on different AI models for GIS, and most recently have been working specifically with LLMs for
3:27 GIS. That's a quick endearity of how we got to - Yeah. History major got into AI for GIS. No, it's funny. One of the devs that we hired at the beginning of the summer is miners in history. So I
3:42 feel like history is actually just a really good major subject that more people need to have in their
3:51 repertoire. So before we get too far into some of this other stuff, to like I want to go back to your billion dollar idea like. Did that just get blown up because the market that you were training
4:01 all this circle they don't just like doesn't even hold water to what's happened since COVID. So then it actually the model worked pretty well to my co-founders credit. We got it working pretty
4:12 accurately. It's just that we found that real estate developers on real estate investors are
4:20 either they work in a very sort of local area. So the work in like a neighborhood of a city and they know it inside and out and they don't need this data or they're more than happy to spend the extra
4:32 money for the sort of ground truth data. You know, this unit in this building rents for this much money. It's like on the lease. It's known. And they're willing to pay up for that. And so there
4:42 wasn't really a middle market where they didn't want to spend, you know, 50 grand a year on this super high-end data, but they also sort of are investing outside of what they know So it just turned
4:54 out that people didn't really want it. And seeing as that was our sort of first venture into startups, we didn't really think much about ask what people want before we built it. We thought, this
5:06 is a good idea. Let's do it.
5:09 And yeah, it just turned out that people didn't want it, but it was a good experience. That's fair. No, I mean, to your credit, you guys didn't just stop and throw in the tally, they're
5:22 epivoted. But that's most successful companies were something else when they originally started, or they had a target in mind, and their thesis was wrong, but their solution had a space somewhere.
5:38 And so I think that's a
5:41 big problem to you guys for continuing and finding where the applications are, because obviously GIS data is literally everywhere And so yeah, I think that's a that's an interesting thing. The
5:53 other thing I wanted to ask about was your it says on your LinkedIn that you were in the fencing club and I've had a very weird like fascination with fencing since I was in like middle school and I
6:03 saw it in the Olympics one time and I was like this is interesting and actually on my commute I drive by a fencing club here in Houston and so it still kind of haunts me of like is this is this
6:15 something I want to try because it looks like it would be a lot of fun. It's really fun. Yeah it's uh I would still be doing it now if I could sort of maybe it's a pro and a con so at USC fencing is
6:29 I believe the oldest club and
6:32 so it gets a lot of university funding
6:36 and normally you need to buy a lot of gear you know by the I forget the terms but there's like a whole get up you get so you don't get stabbed with the sword and you got to buy the weapons and it's a
6:48 pretty pricey thing to sort of get into but because because it's so old and has so much funding. The club buys all the stuff. And so you don't actually need anything to start doing it.
6:59 And so I never actually bought anything. And so now every time I think about getting back into it, it's like, well, I would have to buy all the stuff. And I like fencing that much, but it's very
7:09 fun. If you can find a fencing gym or a fancy club that will let you borrow stuff, it is a really good time. It's very fun to, in fact, find
7:23 the suits But it is as sort of pirate-like, as it sounds. Yeah, it's like a mixture of high-level strategy with just brute force. And I kind of love the dichotomy of those, I think. But no,
7:39 that's awesome, man.
7:41 So let's get into it a little bit. What is bunting labs? And we'll probably talk about Monday as well. But we'll kind of just start from there Yeah, so. Bunting Labs is our company, and at the
7:55 highest level, we're always looking for what the best way to integrate AI and GIS is. My co-founder, he studied AI at MIT, and so that's sort of his specialty, and we think we should make best
8:08 use of that, so we're always looking for different things to do. We started with an AI for digitizing maps. This is, I think, particularly, or has been particularly interesting to geologists,
8:14 because unless you did it very recently, most geologic maps are just like PDFs, but if you wanna work with the data more than look at it, you need to digitize it, which
8:30 is an absurd pain to do because of how complex these maps are. So we trained ourself, 'cause we started in about June of 2022. We trained ourselves a model that can digitize the boundaries of
8:46 polygons, and so it can digitize and retested it, it can digitize the geologic map twice as fast. than if you were to do it by hand. So we started with that, and then we also experimented with
8:57 some other sort of AI for GIS applications. I know I saw you tried it out before we helped on the call. We sort of tested
9:06 making something that digitizes meats and bounds, which we got some interest from land men. We sort of moved away from that. We found that LLMs might not be the best for doing this Like they're
9:18 very close, but it might not be perfect. We've made one for aligning maps. So if you have like an old historic photograph or an old historic sort of drawn map, you can align it in a digital way.
9:35 And then most recently we've been working on LLMs for controlling GIS. So a lot of different things, but really just trying to figure out the best way to bring AI and GIS together. Before we get
9:46 deeper in this, 'cause I know Bobby's got questions.
9:51 Recap with GIS is for people before we really go down this rabbit hole. I just realized that we've been using that, that acronym. Yeah. Yeah. But what it's just geographic information systems, I
10:03 believe, is what it stands for. But I mean, it's tied to anything where you're mapping or plotting, you know, values, you know, or polygons or shapes on a map. And it can get very, very
10:16 technical, really fast with the earth's curvature and all these projections and all this kind of crap. So, but not being super useful. People love maps, love seeing it. They know how to interact
10:27 with it. So, no, it's a huge space. And I'm definitely interested to see like how that integrates with like LLMs. 'Cause like I think we've talked a lot about on here, you know, large language
10:36 models. And sometimes people think that those are a catch-all and they don't necessarily work great with, you know, say, you know, traditional data actually, you know, like our time series data
10:47 and so on curious what the limitations are around. that, you know, current AI LLMs with geologic data, or just our geographic data. Yeah, it's definitely been interesting. So we got into this,
11:04 we started, I think last November, we started seeing what LLMs can do. And I think, like, if you look on LinkedIn and to a lesser extent, you look on Twitter or X or where. There's a lot of
11:17 people who talk about LLMs sort of interacting with spatial data directly. And I think that sort of misses the point. So we work a lot with the QGIS, which I imagine a lot of, especially US-based
11:30 energy professionals are either they use ArcGIS or they, you know, someone on their team uses ArcGIS. QGIS is sort of an open version of that. And. It's gonna say we're gonna have a whole
11:43 conversation about mapping libraries at some point during this. Yeah. So we'll get into that Keep going. Yeah, so we use. We work on top of QGIS because it's all open. And QGIS is all sort of,
11:56 you can program it using Python. And you can also do other GIS work with SQL. And so probably the two best languages that, you know, LLMs know are SQL and Python, I don't know if it's the two
12:11 best, but they're both, it's extremely good at both of those because there's so much of it on the internet. And so we figured we don't need to have an LLM directly sort of interface with spatial
12:20 data. We're working on having to do that. But you can get pretty far by just having it interact with QGIS and sort of turning requests into QGIS code and then having 20 years of QGIS development,
12:37 it works perfectly, having QGIS interact with the spatial data. So there's this sort of middle layer that handles it for you And I think by doing that, it makes it a lot easier to have the LM speed
12:50 be pretty useful. No, it's fascinating. And again, I'm sure like the wonderful people at Esri are working on some of the stuff probably in-house. But I mean, I know that, I mean, there's a lot
13:00 of people who would love to not have to pay Esri for a lot of that. And, you know, so I mean, I think something like this that, you know, can make people, you know, really dangerous and very
13:11 productive, you know, rather, is it probably a better term, you know, within a more free tool. And I guess what you're providing isn't necessarily free You know, it looks like Mundi is
13:21 available to self-host if you want, but I'm sure that's, you know, non-trivial in some ways, but - Yeah, you just put it on a server, Bobby. That's all.
13:32 But yeah, no, I mean, I think it's super interesting, but I think it's a smart way to go about it. We're, again, you've got, essentially what you're saying is like QJS has all the context and
13:40 understands, you know, the requests that you want to ask of it, right? Or how to translate natural language into a request, reprojecting a map or a points or all those kind of things. So, no,
13:48 it's
13:53 very cool. Yeah, any of the processing you would need to do,
13:59 you know, I guess there's a long tail of complex processing you can do, but say you want to find the intersection of two layers or something like that. QGIS already has code that does all of this.
14:12 You just need the LLM to sort of use the right algorithm from it. Which is very easy for it 'cause it's all Python. So
14:22 you don't actually have to have the
14:26 LLM, I guess, work directly with the shape file or whatever. So it's kind of like almost kind of how Texas SQL is currently working where it's like you, you're using an existing software that
14:38 being the database in this use case. And then you essentially train, or you upload the schema and the relationships within that software and then you give that to a model and then it just does its
14:50 thing, generally speaking. I assume that's why y'all went with ArcGIS because of one, it's obviously open source and then two, but it has all of those functions already built into it. So you're
15:04 not having to sit there and fine tune a model to do stuff that already exists. Yeah, exactly. Yeah, QJS has everything we need and it's open so we can, you know, if we were to do this with S3,
15:18 it would, we would have to like go through them and work with them, which, you know, nothing against them, but it would be, it's a lot easier to sort of experiment and move quickly when we can
15:33 just sort of control it ourselves.
15:35 And we can push stuff out. It just makes it faster for us to iterate. Yeah, now that makes a ton of sense. Yeah, so I mean, I feel like, and even to cry if I'm wrong, there's kind of like this
15:46 line of demarcation-ish where I think they kind of blend together, but between, say, like, bunting labs and Monday, or like, I mean, like, so it looks like you've got like some of these suite
15:54 of tools that are built into QGIS,
15:58 AirQGIS, but then you have Monday, which is more of a full platform QGIS of top on built like sounds, built of Kind..
16:06 I don't know if I'm gonna kind of talk about the differences there, and maybe we can talk about, it looks like you've got about four major products, you know, that work within QGIS, and I'd love
16:13 to dive into each of those, so. Yeah, absolutely. I guess on the four products, we definitely have four products on our website. There's a wide range on how well they work. We just kind of put
16:29 stuff out as soon as we think it might be useful, and we have a bad habit of not taking them down when we realize there are problems. But we can get into that after That's a good startup for you
16:40 though.
16:42 Just shift by that. What's the worst that happens? It doesn't work. If you don't ship it, it also doesn't work. So there's, you know, what's the difference? But
16:51 in terms of Moondi versus our stuff in QGIS, you know, around November of last year when we were thinking a lot about LLMs.
17:01 We realized that the LLM Labs, OpenAI and Google, are moving so much faster than we ever could building our own AI and training our own AI, which we've done plenty of, but they just are, it's not
17:15 even comparable, right? And so, and they're so, the LLMs are so powerful that we just need to do. And so we made an LLM in QGIS, and it worked pretty well, but it was kind of janky. And what
17:31 we realized we had to do was, if we wanted the LLM to be the focus, we would need to sort of build a platform around the LLM where we can have more control, sort of. around the infrastructure of
17:43 it. And then if we need to use QGIS, we then call into it. So the way it, instead of having the LLM in QGIS and have it work with your data there, Mundi, the LLM has access to QGIS. So if you
17:56 have two polygons and you need to find the intersection, it then will call into QGIS, run the processing and send the data back, which makes it just a lot easier for us to get the most out of the
18:09 LLM's, which is sort of why we're moving in this new direction. It probably would be strictly better to be in QGIS, just because there are so many people already familiar with those workflows, at
18:21 least for people who are very experienced at GIS.
18:25 But we think that it's more important to get the most out of the LLM and you do that with your own platform versus building it inside of QGIS. But there's also the, what we're finding with Mundi is
18:38 that it's very appealing to people deeply experienced in QGIS. I don't know if we've ever opened ArcGIS or QGIS. It's a pretty daunting platform to get to use. There's a lot of knobs. There's a
18:52 lot of knobs, a lot of buttons. It can do anything you could possibly want, which leads it to being very complex. And we think having our own very sort of straightforward, simple platform, it's
19:06 a map and a chat box It also brings sort of the ability to do spatial analysis to people who aren't deeply experienced with GIS, who don't want to open our pro or open futures. Yeah, I guess that's
19:20 what I was gonna ask. Are you targeting, and maybe it's both, or in some way, but are you targeting, are we gonna be an amplifier for professionals, GIS professionals, that cursor is for
19:32 someone like me or John or whatever, where it's like, we generally can do some of the coding or a lot of the coding, I can do it faster and maybe better by testing around my code that I wouldn't
19:43 necessarily otherwise do and so on, or is it targeted at, because let's just say there's a reservoir engineer who isn't great, doesn't GIS professional, but knows what he wants from it. He's like,
19:54 I want to know what wells, you know, what, you know, line strings, intersect, you know, these counties are intersect, you know, this, you know, DSU map that I, you know, plop on the map
20:07 And it's just making it easier for them to do those workflows. Yeah, I think we're, we're trying to have it be useful for both. Um, and we sort of have now two sets of users. We have users who
20:18 themselves are, um, you know, very experienced with GIS, very experienced with spatial databases. Um, you know, most commonly there'll be database geologists and they could do all of this stuff
20:31 themselves, you know, in QGIS, in ARK Spro, uh, but it's just much faster and easier in our platform. And so that's appealing to them. But then the other set are
20:44 sort of firms where there's a lot of people who are not specifically GIS analysts, but they need to work with spatial data. So you might be like a consulting firm of some type. And you have a GIS
20:59 team who manages a spatial database. And then a lot of consultants who are not GIS analysts, but need to use this data to do whatever it is that they do. And so with our platform, they can explore
21:10 the data more quickly. They can spend less time looking through the database, writing SQL more time, finding the right data that answers the client's questions. And then once they know what they
21:20 want, they can send it to the GIS team in QGIS or in ArcPro. And they can make a nice pretty map out of it. But the consultant can spend more time sort of on the impactful work than fiddling with
21:33 QGIS knobs and writing SQL. OK, very cool.
21:40 So, I mean, I may guess with this, I mean, obviously we're energy bites, I mean, because GIS spans like all kind of industries, but I mean, how much exposure sounds like you've talked to some
21:47 landmen? I mean, I'm curious how much exposure you've had to some energy workloads and like, what does that look like? Yeah. So
21:56 the most direct exposure we've had has been through the landmen. So when we were before we started working on LLMs, sort of controlling the whole GIS, the first thing we wanted to do with LLMs was
22:09 digitize the legal descriptions. Obviously, it's a big problem for landmen.
22:16 And that product is so - It's damn playing it. Yeah. It really is. Like, I was - one of our projects, we pulled in a bunch of public data, and each asset has probably half a dozen to a dozen
22:29 plats of some sort for each individual well. Yeah. And you're like, oh, good It's just map and like for so. you know, for enterprise rag, it's like, okay, I can tag this as a map for this
22:41 asset, but not really doing much else with it until I saw your stuff. And I was like, oh, this makes this very interesting. If I can quickly digitize it, and then now it's digital, so I can
22:53 actually use it on a map if I want to.
22:58 It's yeah, sorry to cut you off, but it it like we have so much paper data in our industry that like, it takes, that's one of the main reasons I wanted to have you on is because there is a lot of
23:11 need for this type of stuff. Yeah, yeah, it,
23:15 the the the Platt digitizer
23:19 turned out to be a lot more complex than we wanted it to be. So it kind of paused it. It works fine if you have a very straightforward, you know, legal description, go 100 feet, turn, you know,
23:34 20 degrees, whatnot. Uh, that was fine. But once you start getting into more complex stuff, once you start requiring other data, so whether the point of beginning is unclear. So the point of
23:45 beginning is like, start at the tree on the lot. I don't know where the tree is. Or along
23:52 the property boundary that belongs to the Jones family. Well, the LM, neither the LM nor I know what that means. And so we can't help with this. And that turns out, as you all know, that's a
24:06 lot of them. And so it just turned out to not be too useful. But we got some exposure to LM in that way. Most of the exposure to energy geologists through. 'Cause been has as
24:17 we trained this map digitization model, it turned out to be most useful for geologic maps. So almost everyone who uses it is a geologist, digitizable maps. And then now again, with Mundi, a lot
24:34 of the people most interested are geologists. just because they have so much data and
24:41 it would be, it's so time consuming to work with, but LLMs make it a lot faster. So a lot of exposure to geologists and some exposure to landmen is sort of what we've seen from the energy sector.
24:54 What are the primary use cases outside of the energy sector or what are your kind of most popular features there? Yeah, so outside of it, it would probably be what I was describing earlier with the
25:07 sort of consultancies where you have consultants who are
25:11 not GIS experts. And, you know, a big one we're seeing is transportation consulting. You know, you need to figure out where to place a bus stop or something. That requires a lot of spatial data
25:24 and, you know, these transit planners are experts in that, but they should be spending their time thinking about sort of the client's questions and not messing around with the GIS.
25:34 And we've seen a lot of interest from sort of GIS managers in these consulting firms who understand that there's a much better way that their clients could be working with the data or their sort of
25:47 end users in the firm could be using the data rather. And they're just not, but this LLM provides sort of a new, a new way to get the most out of what they have access to. Yeah, I think that's
25:59 gonna be the overarching trend with language models across every industry, right? It's a tool that empowers or democratizes data that you already have by letting people who don't know how to code or
26:14 speak whatever language it may be to have access to that. And I really do think that that's like the near term, like impact of whatever gardener says. Oh, I agree. UI and all of them, that's
26:27 like the boots on the ground impact that
26:31 this is having today. Right. I totally agree, there's always like, oh, if the LMM was better, it could help experts do more. But right today, it's good enough that it can help beginners go
26:46 really far. I
26:50 have a history degree, and with cursor, I'm able to make their minor edits, but without it, I would have no idea how to do it.
26:60 I can write a very small amount of code, edit the stuff we're working on.
27:06 Yeah, it just makes it a lot better. And it's not to say that it's not useful for
27:11 experts, very interestingly. So before we started working on boondi, we started working on an LLM inside QGIS. And most users that got are 10, 20 year veterans, GIS analysts, and it's helpful
27:31 for them, I do think that it's far more helpful for people who sort of are new to the industry, sort of want tangentially to it, and let them get into it too. I'm hoping QG has up because they
27:42 don't, they're scared of it. Yeah, if you already know how to do it, it makes you faster for sure, but it doesn't sort of unlock it. Like it doesn't make you now able to do more than you could
27:52 before. And it's good to be faster, but it's better to be able to do more than you could. Now, how about like one thing I'm thinking here is like, all right, let's say you've got web developers
28:02 that, you know, are not GAS folks, but like maybe like I, so, or would they be able to say make a map in Mundi and then embed that in their application 'cause they point to it. And then like
28:13 that's the mapping surface almost like I guess. And I guess maybe we can start segueing into like, how does this integrate say with like a map box or something similar, or is it a replacement for,
28:22 or does it integrate with just curious on that side of it? Yeah, so this is actually something we have been working on somewhat on the side. as we've been working on Moody, our primary focus has
28:33 been on working with spatial databases, but we have been making it so you can embed it in websites. And I think what is most interesting about this as it compares to other embedded map solutions is
28:48 we want it to be very easy to embed a map and then have people sort of edit that embedded map more quickly. And so, you know, you could embed the map in a site or an app or whatever But then you
28:59 could also have this easy-to-use interface to change the data and add new data, you know, just change what's displayed without having to also, you know, go open up cursor, open up whatever it is
29:11 used to code and edit the map that way. It makes it a bit easier for non-technical people to edit the embedded map. That's a great feature because I feel like it's probably because I'm an engineer,
29:25 but anytime I see a data set, I'm like, okay, well, I want to do something with that data set. And then most of the time, the data for the dataset is nowhere to be found. And so you're very
29:34 limited a lot of the time, but that's a, no, I mean, I think that's an incredibly valuable feature as small as that sounds. That's a something that everyone ends up wanting to do at some point,
29:48 right? Yeah. I mean, we, so Mundi, I think, I guess they're all technically open source, but since Mundi is more of like a platform than the other ones, I feel it's different that we've made
29:58 it open source, but we do want people to take it and sort of build on top of it, you know, and make it their own. We've seen some very interesting applications of this already. Actually, right
30:10 after we launched it, we found someone taking Mundi and using it to make like a custom map for like a fantasy world, which is very separate from any sort of commercial application But, you know,
30:25 it was cool for us to see that someone saw this platform and took it and used it to
30:31 make something they wanted to build something. Yeah, build something on top of it. And I guess to that end, I mean, like, does the map have to like fit in this, you know, with a base map of
30:41 the world? I mean, 'cause I mean, like, when I think so, a popular platform in oil and gases, typical spot fire, which is like a BI tool, but it's got really strong GIS. But actually like
30:52 within like say a map chart in there, you can actually have one where it doesn't actually have a base map behind it. And that's actually how you can plot certain points in space, like then create,
31:03 I think people used to do like ternary diagrams and stuff like that, like within there where they'd put an image on the background and then that was almost like the map and then you could place, you
31:12 know things within there within a
31:16 geospatial-ish kind of context. But I guess just thinking about that, I mean, does it have to be a map of the Earth or like a global view or can it be just like limited to a, you know, a local
31:26 region or like you said, even like this is in this fantasy world, like you make up your own geography and topography and so on. Yeah, I think it can definitely be changed. And I think, you know,
31:38 it's very useful. So we default have like a,
31:43 I forget the actual libraries we use, but we have one that's just like an open street map, Google Earth, not Google Earth, Google Maps type, just vector base map, and then we have a satellite
31:54 base map But there's all kinds of things you can change it to.
31:59 One of the sort of, it doesn't yet support it, but something we're looking at is a thimetry, and so I could totally see, it would be useful to have some kind of base map that lets you sort of see
32:11 below the water, which I don't think either of base maps currently would support. But
32:18 yeah, there's a lot of different things you could change it to. I didn't see, I saw this a while ago, NASA uses it and change the base map to the moon or Mars or something. And so for them,
32:30 that's actually a very concretely useful function.
32:36 But yeah, there's all kinds of different applications that you could change it to. You know, I think that also goes to the open source version. If you have something that you specifically need,
32:46 you can just take it and do whatever you want with it. You know, make it your own I just say, and I know a lot of geologists are getting excited when they heard you say bathymetry maps. And
32:58 they're probably licking their rocks right now.
33:02 So, and I guess some of this comes back to the fact that you've built this on top of QGIS, and so then you get to utilize pretty much everything that's in there. But I mean, because now you start
33:12 talking about base maps, but even the different layers and so on. I mean, are people able to connect to like web feature services or web map services and like bring those in. So not yet, but that
33:23 is something we're actively working on. Right now, to bring data in, we only support PostGIS.
33:31 We're also working on supporting WMS and just because we know that this will probably be important for some enterprises, MS SQL and - Yeah, I was thinking about it earlier. Yes, we wanted to be as
33:47 sort of accessible as possible because sort of, we look at this as a way to, you already have so much data, you know, we don't need to help you create more data. We need a way to sort of
33:59 visualize it and work with it. So having all these integrations is important to us. Right now, it's just post GIS, but you know, we're working on all these other places you could get it from. So
34:10 this gives me thinking a little bit. Cause they'd say a WMS is just like a, kind of almost like an image layer, right? It doesn't actually have like the features there. So it's, you can't
34:18 necessarily say to an intersect join or something utilizing that. But in, I don't know if anyone has this or if you all look at that, but like, is there a way for it to almost use like a computer
34:29 vision type thing where it could actually see where this, you know, let's just say, whether like where this storm cell, you know, if I want to know when this storm cell intersects Fort Bend
34:38 County, you know, which again, I'm pretty sure, you know, whether that comm is doing that for me already. But let's just say I want to do it myself.
34:46 You know, could you actually bring in a WMS and could actually see that this red cell, you know, intersected with it, even though it doesn't actually have the feature available to it? Absolutely.
34:55 So this is probably at the frontier of the stuff we're working on. We do think that. So there's, there's kind of a mix. So the new Google models, I think in particular are very good at this kind
35:08 of And we do want to integrate that in. Integrating computer vision is just generally something we really want to do. One of the features we haven't released yet, we actually haven't announced it
35:21 yet. So I guess this is the first time it'll be out there. One of the features we really want to work on is if you have like a SharePoint or some other sort of database of documents that relate to
35:32 spatial data, be able to combine those. So the example we're sort of working with is imagine you have a spatial database of dams and then you have a SharePoint full of dam reports, like inspection
35:43 reports. And you want to know what dams have recent, in the past three months, had reports of too many plants growing around them. You could then just search that in and it could combine this
35:57 technically not spatial data, but it's effectively spatial data 'cause it relates to an actual point and sort of put them together. It's a certain data back to an asset. Yeah, I think once you
36:06 start sort of pushing LLMs their limits. you know, with all this different data, you can do a lot. So whether it be computer vision of the you know, raster data, with the vector data, or
36:20 you know, searching through documents and combining that with spatial data, I think there's a lot of different things you can do. Yeah, I think it would be hugely
36:28 beneficial because I think, you know, say that getting back to geologists, I think we had some different maps where they were mapping like different geological properties across the acreage, but
36:38 like if you were to take a like a raster or a point layer of that, it would just crash the memory, the RAM on your computer, but we would expose it as a WMS. People could kind of see like the
36:48 gradients and so on. But then if you were able to just like publish like the gradient and then the computer vision is able to actually interpret like it had the scale and it could kind of like
36:57 utilize just that image, you know, and identify that, you know, certain things that be extremely powerful where you don't need to load as much data to get the benefit computationally. Yeah, and
37:10 I think that's also a big benefit of both LLMs and,
37:17 you were joking earlier, oh, you can self-host it, it's no big deal. But if you're doing that and you put it on a server or you use our hosted version, which is on the server, it's just so much
37:27 easier than using QGIS. Over the past month and a half, since we launched Moon DIS, that only using that I haven't gotten into QGIS. And I recently opened QGIS again and then I forgot it takes
37:40 three minutes to open and another minute to open up a new project. And it's just so intensive on your computer and
37:49 so slow that I think using these web-based solutions is just a much more pleasant experience. And until pretty recently, WebGIS, I think had this connotation with weak. But if you start getting
38:03 these sort of powerful LLM features, having it be sort of a full-function GIS, you know, The more you can offload from your desktop and put it, you know, in the cloud. It's just so much easier,
38:17 so much more pleasant. What if I want to drag in a shape file or a KMC or any of the, I mean, like, are we there yet with that or? Yeah, yeah, yeah, yeah. We say I did that this morning
38:31 because that was on the demo. Yeah, you can upload a shape file technically right now You would have to have it zipped with all the, the side car files, but you can do it. You can think you can
38:43 upload KMC. I know you can do a GeoJSON, I can upload geotips. Right now, I think there's like a 500 megabyte limit on the uploads, which obviously we're going to make larger, but we figure the
38:58 way we approach things is get it working somewhat and then get it working fully So we have it working somewhat where you can upload smaller J files or smaller Jujasons or whatnot. and then probably
39:11 in a week or two, it'll be upload whatever shapefile you got or whatever, you know, raster file. I mean, half a gig is pretty good. I mean, like for now, I mean, like. Yeah. You'd be
39:23 surprised. I mean, maybe you wouldn't be surprised. We get some, start asking people, oh, how big is it, you know, they say, Will this be a problem? And we say, How big is it? They're like,
39:31 Well, each file. So when we, when I have to send it to the client, I purchase a hard drive, a five terabyte hard drive and then I mail it to, I say, Well, you know, one day we'll get it there.
39:43 But I mean, you're always gonna be at the mercy of, you know, upload speeds and throughput and everything. You know, I did wanna ask Michael,
39:52 where do you think, what's the like current state of the computer vision side? And where do you see it kind of going? 'Cause like I mentioned, when I told this, told my geologists that we were
40:04 having the podcast and he started looking at this stuff, Oh, well, if I give it a map and it's got the scale and it can understand it, like, could it just digitize the full thing for me, right?
40:18 And there are so many of these, like, whether it's plat maps or geological maps or structure map, like there are so many of these where it's, you know, it's a good map, but it's just not, it's
40:30 either digital in a PDF or it's not digital But if it was and it had all the shape files and everything associated with it, it would be infinitely more useful than it currently is sitting in a filing
40:41 cabinet. Right. Yeah. I think it's, I don't know if it currently would work. I bet with a lot of tuning you could get the current offerings to work, but with the rate of progress with these
40:55 algorithms. I don't think it'll be long until you can do that, which hurts as someone who has invested a quite a lot of time making a map digitizer, but I do think that these
41:08 You know, a combination of both the LLMs and their sort of inherent capabilities. So the Google models, I think, are specifically good at segmenting things.
41:20 But there's also sort of the Facebook segment and anything models. I suspect they're working on a better version of that. But even as it stands now, it's pretty good. So I do think, I don't think
41:31 right now you could just upload something and with a mixture of segment and anything and Gemini, you know, get out a full vector representation. But I do think it's not long until you could. But I
41:45 think what's more interesting about this is whether or not it's just sort of the computer vision in the LLM or enriching the
41:55 vector files, uploading the vector version of the map and uploading the raster and having it automatically populate all of the vectors with the data from the original map. 'cause that's a whole
42:06 other process after you trace it, you know, is adding all the other information. And I do think LLM's very soon will be able to basically automate that part. Yeah, no, I think that's, the
42:18 computer vision stuff is just so wild and it's moving so fast and it's getting so good at a lot of things that, yeah, I absolutely agree with you that that's gonna happen at some point
42:31 and that's the craziest part is like, today is the worst, they're gonna be at it and they're still pretty good and so, yeah. Yeah, I'm almost certain that GBT5 is coming out this week.
42:43 Definitely by the end of the month, but I'm pretty sure it's this week and I just, I can't wait to see how they do with that. With their new, I don't know if you saw, they released an open source
42:55 model. I deployed it this morning. And really, we tried deploying it and it crashed my co-founders computer and he has a pretty nice computer. Uh, but I think just at least in theory that they've
43:06 made something open source that that that is that good that I think GPD five will be crazy. Um, and it's, I imagine right after they do that, because it's how Google operates, they'll release
43:19 something that's super crazy. Uh, they're not fast enough to do it themselves first, but they'll eventually get there with something really, really good. Um, you know, anthropic is moving fast.
43:29 I think all of them are moving so fast That it's just, it's hard to stay on top of it. For sure. Yeah. No, I, uh,
43:39 I got it. I got the 7B model working on O lama on my laptop. I did not, did not try for the, I know my laptop won't run the 128 version, but it is. Oh, yeah. It is on Azure studio. I saw it's
43:53 on grok now. Um, everyone is getting it updated on their lists and stuff. So I haven't really messed with it that yeah, I do think that's where It's most powerful is for enterprises that don't
44:08 want to use for their data security, they're either uncomfortable or there's just some policy that won't let them use OpenAI or use, you know, anthropic or whatever, being able to use your sort of
44:23 Azure AWS or GCP account and just put the 120B on it and have an O4 mini level LLM running locally, I think that's where it's most useful. And I think that there's a lot of applications where that
44:38 will, where before someone would have said, you know, an IT person would say, I would love to use this, it seems useful, but we just can't. Now can say, actually, maybe we can, you know, we
44:48 don't have these same security issues as we did before. Yeah, no, self-hosting is gonna be a very fascinating thing to see how that plays out, especially at the enterprise side, moving forward.
44:60 No, I mean, I'm just trying to think through
45:04 headaches or heartache has been on the on the GIS side. I mean,
45:09 it's
45:11 yeah, I think just like a lot of the reprojection, I mean, you get things coming from like multiple different, you know, coordinate systems or you get this is in feet. And I want to convert it
45:18 to lat longs and vice versa. It's just so many, I guess, you know, because it's kind of maybe beaten into a horse, but it's just I guess, you know, QGIS handles that, you know, beautifully,
45:29 because that's what it does And yeah, QGIS handles it and LLMs are specifically good at,
45:39 like, keeping full track of what needs to be done. And so, you know, in an example that I use, I say I have, I guess the projection is in feet and I say buffer this line to kilometers and it
45:51 just knows it has to be rejected to something that uses kilometers and then can, you know, do it correctly, or it doesn't have to be projected, but it knows it has to do this sort of calculation
46:02 to turn two kilometers into feet. It's just very good at keeping track of everything that needs to happen to keep the GIS work running smoothly. So that plus, you know, as I said, QGIS is perfect
46:13 at all this stuff. I think it just makes it, it's a perfect combination. It keeps track of everything that needs to be done without you having to do it. Yeah, and I think that there's another use
46:23 case. And again, I'm not sure how, I guess it would just, it would just start cranking on QGIS and it would depend on how fast I could process it all. But like, I know one thing that I know, a
46:31 ton of people have always had to do is like, all right, what's the spacing between one well, to the nearest well on either side of it? But then, you know, giving it prompt, I guess you have to
46:40 prompt it in such a way that like, take each well and find any well that intersects it, you know, at least, or like, you know, if you find the nearest well to the east or west of it, that's,
46:50 you know, overlaps by 50 or more and all these things. But like, I'd be just to see if it could actually output, you know, almost a table of that iterating over a data set like that. 'cause
47:04 there's a lot of people that have their Python scripts or SQL spatial or whatever it is that's done that or tried to do it, but be curious to see how well it handled it or how long it would take.
47:16 Yeah, I think it would probably be good at it. Something that we have recently fixed, it now is able of sort to do these multi-step sort of more complex reasoning tasks. So it can write complex
47:30 SQL and then run a series of geo-processing algorithms. So one of the examples that I was testing was find how many of these polygons are within two kilometers of this line. And so it knows, okay,
47:46 it needs to sort of filter for the right line, and then it needs to buffer it, and then it needs to find the intersection of the buffer, and it's able to sort of keep track of each step. So I do
47:57 think it would be able to do it. In terms of specific oil and gas work,
48:02 what I think will be interesting is seeing how much you need to prompt the LLM and how much it just understands, okay, this person is doing, you know, a request about oil wells. This is
48:14 specifically what they mean if they say how much do they overlap, versus how much do you need to give it more specific guidance of, here's what I mean by overlap, here's how you should sort of
48:24 approach it, and then take it from there. Yeah, I think the problem is a good thing to bring up here, and that's kind of where I want to go with it too, is like, right, how much, like, I mean,
48:35 imagine you've got some system prompts behind the scenes that kind of this all gets feedbacked into so it has a ton of context, but then do you also allow maybe like a customer or a user to upload
48:46 like their own, like, you know, say, you have cursor files and different folks have it now, where like they get upload their own custom, you know, system prompt that enhances what you have,
48:53 and then it could make their curing better or easier as well.
48:58 We don't do that yet, but it is something we want to do. We actually, and I'm very curious to hear what you think, cursor for oil and gas looks like. We take a lot of inspiration from cursor.
49:10 Something we do want to add is custom prompting. And so we do have a very long, complex system prompt that we just keep adding to and keep iterating on. And then obviously the user can put in a
49:22 prompt per request. So with each thing they want to do, they can add more to the prompt But we do want to add something that lets you always have this be part of your prompt and be sort of more
49:35 custom to what you're doing. But I'm curious, as someone who uses cursor and is so in-depth with oil gas, what do you think cursor for - either cursor for GIS or cursor for the oil and gas industry
49:50 it looks like? It's a big inspiration for us Sorry, John. He's kind of sort of. building it. I'll take it. I'll take it on the high level. Bobby probably has a lot more experience taking it on
50:03 the GIS specific piece. But, you know, to me, it's an interesting question because there are so many incumbent softwares out there. I mean, we have multiple data or document management softwares.
50:18 We have multiple reservoir engineering softwares, multiple production softwares, multiple, there's all this stuff, right? And so, I personally am curious to see how much of it is kind of
50:30 architected similar to how y'all have, where it's like, leave the infrastructure that's there, and let's just talk to it in an easier way for most users. The experts can still go into their tools
50:42 and use their stuff. But Susie and accounting, who doesn't know shit about GIS, just needs this one little thing so that she can get her job done, and she can go use that, right?
50:55 The hardest part with our industry, which I know it's not unique, but it's just the diversity of the data, right? You have everything from, you know, there's so much paper documentation. That's
51:07 what we focus on at Collide as energy enterprise rag, right? So basically just being able to search through your entire document set and find the answer that you need as quickly as possible. And
51:20 then we're starting to build some automations and stuff off of that based off different kind of repetitive tasks that engineers have to do on a daily basis. But then you've got, you know, that's
51:30 just the text data, right? There's embedded within that text data is images, diagrams, plots, plats, all kinds of maps, all this other, you know, rich, rich text type stuff. And then we've
51:44 got all the machine data. So you've got just giant repositories of SCADA data that have been running for decades that have tons of information in them, and no one knows, you know, that's where.
51:55 our industry, in my opinion, really started kind of adopting the AI stuff is around long-term time series, preventative maintenance, or failure predictions, or things like that type of stuff.
52:08 And this, to me, is just another iteration of that. It's like, hey, we needed actual AI and machine learning for understanding all of this numeric data because as humans, we can't understand all
52:21 of it because there's so much. It's the same thing with text. It's like, we just have a shit ton of text data. And the easiest way for us to use that data is by layering AI on top of it. So we
52:32 can get to the answer as quickly as possible.
52:36 So it's, it's tricky from that perspective. Then you've got all the accounting systems and all the other fun crap on the back end. And so it's, you know, I don't know that there's going to be
52:45 like one, you know, one stop shop, at least anytime soon.
52:53 But I do think there are gonna be a lot of these tools where it's like, hey, we integrate with all your geospatial data and anyone in the company, no matter how skilled you are or not, can use it
53:04 to get answers. And I think it's gonna be the same thing on the tech side. Hopefully, I believe we'll be the company that does that. And then there's already been lots of companies doing this on
53:15 the true data machine learning side for a decade plus now And so it's gonna be very fascinating to see how all of this stuff starts interacting and layering on top of each other, especially with MCPs
53:28 and all the other fun stuff that's coming about with this. And so, yeah, that was a long-winded way of saying nothing, but I'll turn Bobby. Bobby probably has a lot more thoughts on the GIS
53:41 specific piece. Well, I mean, I don't know if he was even asking GIS specific, but just kind of how an oil and gas like cursor for oil and gas. And I guess, I mean, I don't know if we're
53:51 speaking about cursor specifically or just kind of that concept of being able to ask questions of it. But I mean, I'm thinking from the software side, like I'm a huge proponent of like, I get an
54:01 IT people will probably has will explode. But like, I mean, I mean, I think every, you know, tech for, you know, Python, savvy reservoir, you know, our virtual engineers should have cursor
54:14 or something similar at their disposal We're seeing clients do that already. And like with their drilling and completion engineers, just giving them cursor so they can write simple Python scripts to
54:26 solve their problems. It's like, it's not production code. It's not getting public. It's not enterprise grade, but it solves my problem. And I think I completely agree with you, Bobby. That's
54:37 just, I mean, Seth, Seth, I'm loose. I mean, whether it's, I mean, even like a, it's a production tech river, like, I mean, like, we had a lot of use cases where we just We're seeing a
54:44 simple crud app and it's like. All right, Power Apps is fine, but it's kind of paying the ask, but I mean, why can't like I just, you can literally go into cursor right now. Probably I could
54:52 tell it, I need a simple crowd app that works on mobile and desktop. Here's the five fields that I need. This one's a date. This one's here. Here's a list of wells, you know, put this in the
55:03 drop down and, you know, so that you have some data validation on it and, you know, the values in this field can't be less than 1000 and greater than 10, 000, you know,
55:12 whatever it is like, and it would generate that immediately And it would probably, you know, use Bootstrap and it would look fine. It'd be better than anything, and it could do that in 10 minutes.
55:22 You know, and I could go on level, we'll do that, and it writes it right into, you know, Superbase, which is Postgres and we're off the races. But, I mean, to a whole nother level, John, I
55:31 mean, I'm thinking like, why couldn't, you know, so John and I worked together at a company that was streaming, like, basically pressure data for, like, for a frack side of it, when we did a
55:41 lot of what we're called Defits or Mini-Fracts, John, why couldn't we have taken, you know, any or a bunch of those defit papers from SPE provided those as context to in LOM and say, you know,
55:56 write this in Rust. So it's really damn fast too. And like, you know, if I give you columns for basically date time and pressure, you know, write me something that identifies when closure
56:09 happens based on these three methods. Like, I mean, I'd be very interested to see how it goes. I guess I could go do that here after five minutes we're done.
56:19 But it would just be really interesting. It's just stuff like that, where you've got all this. And again, John, I've pulled in and probably a lot of technical papers that they have access to and
56:28 just, you know, transcripts, I'm adding transcripts are a huge thing too. Like you have all this, you know, these conversational data that, you know, happens on the morning operational meeting
56:37 and it's like, and then you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you
56:38 can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that
56:38 you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see
56:38 that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can
56:38 see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you can see that you You know, someone said that, you know, the gym
56:40 Bob 5 H, you know, took a pressure kick when they were drilling through it, but then does anyone remember that five years later when they're drilling a well adjacent and so like not in the notes,
56:49 not in the procedures, not in any of the drilling reports, but it was said in a conversation. And yeah, I'm going to get on a soapbox about this for a minute because our industry needs to do this.
56:59 It's like, I don't care how concerned you are about the security privacy piece of this. Yes, that's a thing But like all of the information that you're trying to retain from all these people
57:11 retiring and, you know, leaving the industry gets said in those meetings. It's, it's not written down anywhere. It's not put in a book. It's not put in a, in a report. It gets said verbally.
57:22 And so it's like the technology is so good right now with text, audio to text that it is a shame that more people aren't using transcript data for a lot of their, uh, their stuff because it is just.
57:36 So simple, honestly, like it is a couple of lines of code. Yeah, but I mean, I guess, you say, do you take that example? I've talked about that ties it back like, you know, to a geospatial
57:47 aspect. Like I mentioned a well, we know that this well has a lot longer, even a deviation survey that is a geospatial, you know, it's a polygon or a line string, whatever. And then we also
57:58 actually have a three dimensional aspect because if they said, well, when we hit, you know, this, you know, ash bed or whatever it is at, you know, 10, 000 feet, you know, like, now you
58:10 actually have a three dimensional geospatial aspect to this, you know, that you can go right from, you know, a verbal text thing right into something that's usable data that is super important,
58:20 you know, potentially years on the line. Yeah, yeah, that's, um, I mean, I totally agree. I think, as even if you have, I think people, it's not to say they should be concerned about the
58:31 security of it, but there's so many ways you can do any of this stuff without you know, running into crazy security issues. No one is saying, stream it straight to, you know,
58:42 put it on the internet. There's that, there are ways to do it. Put it in your public S3 bucket. Yeah, I mean, you know, I have a friend who works at a company that is very concerned about their
58:55 data. And I think there was like a fight over whether or not they could use Microsoft Copilot. And to explain to them, Microsoft makes so much money. They don't care what we're doing. They're not
59:04 gonna take it And they're gonna keep it safe. Well, Microsoft has its own problems now, I guess. But, you know, they don't care what we're doing. It's perfectly safe to use, you know, get up,
59:13 go pilot or whatever it may be. Or even if it's not, there's so many local model options that, you know, if you're drilling for oil, you probably have enough money to get, you know, a laptop
59:24 with a GPU on it.
59:27 It's, you'll be fine. There's just so many ways around. Get out of here with your logic is crazy, but. also interesting, you know, what you said about the voice data be geospatial. I do think
59:40 that's another thing that the elements will unlock is there's so much. If something refers to a point on the earth, it should be geospatial. It's obviously it's not a shape file, but that that
59:51 description is geospatial refers to. It's all related refers to a point. And I think LMS will make it
59:60 a
1:00:02 lot easier to work with that. Oh, man. Lots of fun stuff here.
1:00:08 Just real quick. And I'm not I don't want you to say anything that's going to get you in trouble or expose any kind of IP or anything like that. But what is what does y'all's kind of stack look like?
1:00:18 What what cloud are y'all running on? And what are y'all running under the hood? That's a good question
1:00:26 So again, you can say I choose not to answer because I would I would happily answer we um uh it's just that I don't write in any of our codes, I don't know, a lot of these answers. I will say, I
1:00:40 do know we use a lot of OpenAI's models. We've tried a lot of the different ones, and we've just found that the OpenAI models are the best ones for what we're doing.
1:00:49 But beyond that, I think we're pretty, we've built Moon-D on a lot of open source geospatial technology, just because it is itself open source, so it kind of makes sense
1:01:03 But I think the most important sort of decision we make in a day-to-day about our tech stack is the model. And right now, we think that OpenAI's reasoning models are the best.
1:01:15 But yeah, we're always looking at what's new. I imagine when GPT-5 comes out, we're gonna look at that. And then when anthropic and Google put out their own versions, we'll look at that too. But
1:01:27 for now, it's OpenAI. Cool Oh, that's. it's. It's funny because it's like, I can ask that question and you can tell me you're using OpenAI for the models, but then it's like internally, I also
1:01:40 know you're using a bunch of other stuff, just like we are, right? And so it's funny to
1:01:47 just to see what everyone ends up using GPT for the production stuff in most cases. Yeah. It's very good. It works best for, I mean,
1:01:60 working with these geospatial queries, they're pretty complex. And so we wanted something that can sort of go from how many, just keep using the example I've been using, how many of these polygons
1:02:11 are within two kilometers of this line. That needs to be broken down into a series of steps and then sort of handled sequentially. And
1:02:20 it's not like the most simple, it's not just like a question and answer thing.
1:02:24 And we've just found that the GPT models are rest of that. Well, cool Let's uh, we'll go ahead and jump into the this. the exciting speed round to close this out. But definitely appreciate you
1:02:37 chatting with us about this stuff today. Michael, it's been a lot of fun. It's
1:02:47 been great, thanks for having me, it's been awesome. So at the end, we just pepper you with random questions, just short answers, nothing crazy. But what is your favorite cloud provider?
1:02:53 Favorite cloud provider?
1:02:56 AWS. It is pretty ubiquitous. I always ask people the favorite restaurants in Houston, but all right, I've actually haven't been to LA, but I know it's on our
1:03:07 list of places to go, especially if we're gonna take the kids to Disney World, but our cuisine land, but yeah, if I
1:03:14 get out to LA, I mean, I'm sure there's a lot of things that what's your best under the radar restaurant in LA? Under the radar restaurant in LA, easily via as tacos in Highland Park, the
1:03:26 original location is really good I can't recommend it highly enough. And like the guy that operates it is awesome. It's just a great place. I would say, if you go to LA, you have to go. All
1:03:39 right, sweet, noted to that.
1:03:42 What is your favorite non-open AI model? Favorite non-open AI model? Claude Forson it. That's a good answer. Is that your cursor default? I think it is, I think we use that pretty heavily I know
1:03:45 if a lot of people at
1:04:03 companies that are not at all cost-conscious use Opus, but we found that Sonnet is almost exactly as good for a fraction of the price. Same.
1:04:18 I'll say with the kind of the software questions, what's your favorite place to go on vacation? Your favorite place to go on vacation?
1:04:24 Recently I've been loving San Diego. I think it's, especially now that I'm in San Francisco, which is a lot colder than the rest of California. San Diego is just such a nice, pleasant town.
1:04:35 Everyone's happy because they live in San Diego.
1:04:39 It's delicious food. It's just such a nice place.
1:04:45 Yes, San Diego. What is the coolest thing about being a YC company? I'd say the coolest is how helpful other companies are And if you, if another YC founder either works on something related to
1:05:06 what you're doing, they'll talk to you about it. If they know someone that might be useful, you ask, they'll intro you. It's just such an unbelievable network. It unlocks a lot of opportunities
1:05:19 that I think just being a startup, a general startup that would not be available to you What's your favorite mapping service, not named Mundi? They were mapping servers on their movie. It's a
1:05:35 huge discount. I think you just, it is astonishing how feature-completed it is and that more people don't use it. You know, it's not to say that S3 doesn't offer things that you just, you know,
1:05:51 has, they
1:05:53 do a lot of other stuff. But if all you need to do is like, do make a professional map, you just does everything and it's free, you can just download it And it works on Mac. That highly
1:06:03 underrated aspect of desktop software is that I can use it and I don't need to go out and buy a PC. That's a good answer. Yeah, yeah, I don't think about that because I basically get railroaded
1:06:16 into and windows for the most part. But because of that, you know, because it's pretty much everything else, just all the software and oil and gas from those part runs on.
1:06:26 on Windows. I mean, I remember actually the laptop that I used for consulting. I bought originally as a, it was a Dell, it was, it had a Ubuntu on it was like, I thought I was going to be cool
1:06:35 and get a Linux laptop. And then I realized it was useless for most of the things I was actually going to consult on. And quickly switched it over to Windows. And here we are.
1:06:47 That's, that's awesome. Michael, I appreciate it. Where can people find you? How do they get in touch if they want to reach out or learn them, learn a little bit more? If you want to see what
1:06:55 we're doing, moondai, m-u-n-d-iai. Is there a
1:07:02 landing page? You can email me at michaelbuntinglabscom. I
1:07:06 guess that's b-u-n-t-i-n-g-l-a-b-scom. You know, we're bunting labs on LinkedIn. Michael Egan on LinkedIn. If you want to reach out to my co-founder, CTO, I'll ask technical questions. I'll
1:07:19 give him a shout out Brendan Ashworth on LinkedIn.
1:07:24 Yeah, we're pretty easy to define pre-responsive email,
1:07:28 you know, reach out with anything you want. Perfect, man. We appreciate it, guys. Check them out. I know we do a lot of mapping and like I said, I came across these guys just on LinkedIn
1:07:38 through my AI perfected LinkedIn algorithm and
1:07:44 yeah, I wanted to have them on the show because I thought they'd be interesting and potentially valuable to you guys. So I hope you all check them out Mike, Michael, thank you again, Bobby, as
1:07:54 always. Appreciate it. I'll see you this afternoon. Thanks guys.