Energy Bytes

0:00 Welcome back, everyone, to a new year, a 2025 edition. Yeah, first episode of 2025 and during Snowpocalypse week of a few stint, which was not in apocalypse, which was nice. Yeah. For once.

0:16 Much different experience than. Yeah. Very different. It was what was the name of that one? I don't even remember. I don't remember. Nobody lost power. Nobody wants to remember to pipe certain

0:27 tags. No powers lost Kids had fun. I know this was the first time I just went ahead and closed the master valve to the house and just drained all the water. And I don't know why I didn't do that

0:37 earlier because it is so much easy. Like there's no worry of like waking up and being like, shit, the water's not working. It's just like, hey, you've drained all the pipes and you can't freeze.

0:47 But how you been, Bobby? Good. Life's good, man. Awesome. We are here today. I'll let you do the introductions. It's y'all are, y'all go back further than me, but we're here today with Brock.

0:58 How you doing, man? I could be here thanks for I'm glad We've we figure out a time to get everything lined up and all that the holidays were crazy for everyone so Glad everyone's back I'm glad to be

1:10 back on talking to everybody yeah not one hundred percent yup it's really cool for for me and I think a help from rocky but yup he and I put both a need the blue jays from a needful high school down

1:22 south of Houston so I pose for about seven or eight years ago we ran into each other at SB Conference and realized that have very similar Yeah industry backgrounds at least er Yeah Yeah things that

1:33 we're interested in so Yeah you have what one one grade ahead of me and need needful one or two Yeah cause you are in Earnest Great my Brother's Great Yeah Okay Yeah Yeah small high school and now we

1:44 both somehow ended up in Tech and analytics and it's Awesome Yeah I was really cool the crate like people oil and gas is such a big global industry but it is so small and the reality of it because of

1:56 like things like this right like just never know you're you went to school with that is also in the industry or whatever, but yeah, you know, Brock, where where do you work? What do you do? Tell

2:07 us tell us everything. Yeah, sure. So the CEO and founder of Wiserock upstream oil and gas analytics software company.

2:17 So just the overview of our software is we're focused on the production engineering side of things. And what we do is bring in all the data that a production engineer or an operations engineer needs

2:29 to do their job. Into one lightning fast interface. So we pull in all the forecast data from like an areas or combo curve, all the production allocation data from like a pro count or inertia, you

2:40 know, well tests, comments, etc. All the well history from like open wells or wellies. And then every single point is SCADA as well. And the company and we can do those implementations, you

2:53 know, crazy fast because we've kind of built a data warehouse as code. We've got standard connectors to all those things. We bring it all into a lightning fast user interface even, you know,

3:04 across over 10, 000 wells and

3:08 we then enable people to talk about what's going on so they can notice a blip in production with one click, tag their friend in it, say, Hey, this is going on, you know, can you look into it?

3:20 That person gets an email with the exact image that was on that person's screen They can click one button, that other person, and get back into Wiserock with that same view. And so it's interesting

3:31 that we've got all the analytics, you know, we've got all the data vis, but that communication has become a big thing. And then we also allow the production units to set targets on their wells.

3:42 So a big problem on the production sharing side is that you're, you know, you're trying to do exception-based surveillance and a lot of times, you know, comes in able to try to do things that are

3:50 really fancy, but really, if a well is losing production, then you should probably just look at it and be a good starting point. And so the problem is you don't have a good target. Um, and

4:01 generally things like rolling averages don't work very well. So we've made it to where you basically just have ARPS decline curves going backwards and wise rock that, you know, you can create a

4:11 break point in the production history when there's a frack hit or a work over, and then you can just drag it to, you know, adjust your decline and be factor, et cetera, very lightning fast, kind

4:20 of create a history of all the things that affected the wells, steady state production. And then importantly, it's accurate, you know, down to the nearly barrel level for a while in gas. Um,

4:29 now as a production engineer, you've got, you know, very, very good exception based surveillance across, you know, tens of over 10, 000 wells. Um, and then we pull in the, the variance or

4:41 the downtime information from whatever systems it's currently in as a starting point, uh, which is never, you know, capturing all the actual variance. So it's usually 50 or less of the actual

4:53 losses are being coded. Um, and oftentimes that's incorrect. you just click and drag across that and so within a within a couple of weeks of using wise rock these companies have for the first time

5:05 an annotated view of every single barrel of production they've lost every single event that has happened and they're talking about it and then they continue to use up and then we also have a workflow

5:15 to compare those what we call capacity curves back to the reservoir team's forecast which is the only versions truth that everyone else cares about so they can tell them by exception hey you know

5:26 these PDP cursor off like if you want to next time you're an aries or whatever you know you can update them to match and then you'll see those kind of merge book of month or two and they'll start to

5:36 diverge because the production shares are looking every day and then they'll have another meeting and so it's kind of exception base surveillance exception base forecasting and and communication I can

5:48 say I feel like the communication piece is one of the big things that most platforms historically have kind of missed the mark on right beckett like Oilfield operations is so complicated because there

5:60 are so many people there are all of these like the guys in the field are doing one thing and then the guys in the office may or may not know about that in real time or a week later like there's

6:09 there's so much kind of areas of communication that can happen in the field that it's very hard to grasp into one place but if they are all in one place then you have a actual true understanding of

6:21 what's happening right yeah like the data without context is no good generally and then the context with our data is also not as good but together they're much better yeah I mean and you know I think

6:32 that we had a one operator we love talking about is Endeavor that you know they got uh they get acquired by Diamondback and they were one of our earliest customers and so you know they've been very

6:42 public about the value of Wise Rock and or you know we can talk about their experience but you know we we don't um with even with an early version of the product, you know, they're, they're getting

6:54 the point where there's literally, hundreds or over a thousand conversations and Yshrock every single month. So at the top of these, you know, well histories, you see every single thing being

7:04 discussed and it becomes like a digital well file of everything going on. Um, and so those conversations were already happening, but it would be in an email or somebody had to go to five different

7:15 systems, take screenshots, email it to somebody else. That person like thinks that they want to dig more into that data. Now they have to try to go on those same systems and pull out the same

7:24 thing for you and then say, wait, no, this other trend you missed. Well, that ain't going to happen. Yeah. And so it just turns out that we didn't need all the features of platform like Slack.

7:35 The reason why Wise Rock has organically taken off for communications is because we remove all the friction of having the data as context for the conversation. And so now it's the easiest thing to,

7:49 to talk about, you know, actual production trends. And, and then the beauty of it is, you know, engineers are always changing between Um, well, now that doesn't matter because you have the

7:59 full history there forever and you can attach files and, you know, we can do everything there. Right. And so, um, that's been really cool to see. And, and what we've done is intentionally

8:08 avoid anything with, um, AI or even, you know, any type of data science algorithms for now because, you know, we found that the real problem was really just data integration So it's a co

8:21 visualization and communication between humans. Well, the point that you mentioned, right? Like, Bob A sees the problem and emails it to me. And then I've got to go find exactly what he was

8:32 looking at. Like, I've never even thought about that, but it's so true, right? Like, if I'm doing a sequel call on my computer and I don't give you the sequel call, you're not going to know

8:40 exactly what the hell I'm looking at. If I just tell you, you know, hey, it's on this plot or whatever, like, or in this dataset. And so even that is an extreme time saver I can have it even

8:50 like some like the I think you're to your to your point, you're not doing any of the data science data science or AI stuff yet, but you're laying the foundation because I mean, the foundation for

8:58 any of these is quality data, but like having like say consistent tagging or brush it, you know, that you can mark these areas and say, this was due to a frack head or this was due to this type of

9:06 this type of workover and depending on how specific or how, you know, you know, I don't know if there's standard buckets that Wiserock has, but then I'm sure people can create their own right now.

9:14 We let them use theirs because that that would be like, basically, there's no change management of the Wiserock that doesn't replace other software It doesn't make them change their processes. It

9:24 just makes the stuff they're already doing easier. So we don't - we'll match whatever categories you have. Now, very often, when you see it in Wiserock, you're like, oh, we've got five

9:33 categories that are 80 of our losses. We should probably break those down to be a little more granular. And then the 50 other categories are used like four times a year, so we should probably merge

9:42 some of those. And we can do that immediately for them

9:46 very easily. And so that can be cleaned up but.

9:51 But Wiserock is almost doing that, you know, you always hear what the 80 of your data science is, the data of cleaning and data munging or even tagging, like, yeah, you guys are basically, you

9:60 know, on top of providing value, you know, just with the descriptive analytics and the communication, like providing the bedrock for doing the advanced analytics. And I'm making talk a little bit

10:09 about like, I mean, people can do a ton in the platform. Yeah. It's been a couple of years since I've seen it. But then even like, all right, you also have this database or data warehouse, to

10:17 your point, that's available for people to take that data into their own, whether it's spot fire or different places and probably enhance it downstream too. So just kind of curious how that works

10:26 or how you've seen it work. Yeah. I mean, you know, sometimes, you know, if you probably heard what I just said, it'd be like, yeah, but I have spot fire and we, like my background started

10:39 Wiserock as a spot fire consultant and Wiserock would not exist without spot fire because, you know, I was sitting there with operators building things. always related to production optimization.

10:50 My background is patrol engineering. I worked at an operator for six years before, moving over to, and while a gas software company was really big, got to learn product management, big data,

11:01 cloud stuff there. I still, again, focused on production optimization type problems, and then started Wiserock doing consulting. And so, I mean, my tools were SQL and Spotfire, and even Excel

11:16 is a UI back to the database for, you know, easily munging lots of data. I mean, it's a good user interface when you're dealing with tables of data. And if you do it right, you know, you're not

11:26 storing data in Excel. So with that stack, which I was, you know, not at all a software engineer, my background on software engineering was a 90-day software bootcamp

11:37 on full stack engineering, but that just means I know enough to ask good questions. Well, now, Chachie, PT, I can really just run the rest of that. But I mean, even as like super beneficial

11:47 'cause you're really dangerous like in the text areas. Yeah, yeah. But like that, so the thing is most people use spot fire as like dashboard where you end up with like 10 tabs. And the way that

11:59 we would try to build it is, usually there's one, but you're using controls, you're using your text area, you're using your iPad and Python and your scripts to make more of a data product

12:14 that you're thinking about the workloads and trying to simplify it, simplify it, simplify it. And it's usually like giving you insights and trying to maybe write back to the database. But what was

12:25 interesting is by building those things,

12:29 man, you can do a lot of stuff, but then you're stuck sometimes because you can't write back to the database very easily. And especially in the production space, like you definitely can't like

12:36 drag it

12:38 through. And really, you can't drag a curve with any of the open source database libraries. Either you'd have to build it like with D3 or something,

12:46 So what I realized, and I told the operator, I was doing consulting with it at the time, full time. Like, hey, I've been building this stuff for you, y'all are excited about it. But y'all

12:56 still can't answer the question right now. They actually came to me, they said, Hey, our CEO said we need some metricsfor next year for the production engineering side thought we And.

13:07 it'd be good to say let's reduce our production losses. How much do you think we should, what do you think about that? I was like, well, how much production are you losing right now? They're

13:17 like, we can't really tell. 'Cause we don't have a target that is reliable. And then it was like, ah, but we need that answer because we're supposed to say how much we're gonna reduce it. I was

13:25 like, well, maybe you could just say you're gonna figure it out first and then reduce it by, you know, 5. And I said guys, but there's this thing that I could build that I, you know, but I

13:35 don't have the skills to build myself. You mind if I go down to 20 hours a week and I'll develop it with my own machines, my own developer. And then if y'all want to, y'all could license it. And

13:44 I hacked together an early version in D3 that all it was was a plot. It was a single plot with a fake production curve and it was at a client curve that you could click and drag it up and it would in

13:56 real time change the curve and you could create a break point and then do it again. And I was like, this is so fast and easy. You can do this for a thousand walls.

14:07 We have, do that for a thousand walls in a couple of days and then keep it up to date by exception very easily And I said, this is the missing piece that would then feed all the other work. And I

14:17 showed that the variance as well. And they're like, yeah, sure, go down, you know, go work. You know, let us know how that goes. And then four months later, that was actually four months

14:27 later is when they came and said, we're trying to do this. And I was like, remember that product? I said, I wanted to work on it. It's like, it would do this. And they're like, oh, that's

14:34 awesome. And they're like, oh wait, we had to figure out the cost. You know, it's so, so then we figured it out. They didn't actually become our first customer because they announced bankruptcy.

14:44 uh, two days before they signed that contract that we had fully negotiated. But you had some start of life, right? Early luck or lack of lack thereof. I don't know. But anyway, the, the, the

14:55 thing was, is that, okay, there's this gap between tools like spot fire and power BI and those systems of record. And so what we did is we built wise rock to be the integration super fast with the

15:08 best modern data stack technology that, uh, I mean, definitely when we started and now very, very few operators are using and then, um, and you know, it's even, you know, it's even snowflake

15:20 isn't, you know, not the right way to do it and I can talk about that. But then, um, uh, the, then we actually white label white, uh, spot fire as a separate component. The core office all

15:32 built custom in reacts, but we have a kind of an analytics panel where, uh, you know, people don't know this usually, but TIBCO is really optimized for selling to. independent software vendors

15:43 like us too now, that's kind of I think where they see their growth with Spotfire. And so what you can do is we can rapidly build anything in Spotfire and they click a button in Wiserock and it pops

15:55 open that dashboard on the right monitor and side of browser on the right monitor, on the left monitor is Wiserock. But we made it to where they're tethered to where you can have a shared document

16:04 property. It's especially named document property. So the full state of our app, it's like, what well are you on? What time range are you looking at? What layout do you have shown? Are you

16:13 clicked on a variance of that right now? What wells are you filtered to? That is available in 300 milliseconds every time you change it to your spot bar file. And then also in your spot bar file,

16:25 you can control those. You can say, no switch to this well, no group by this, no filter to this, no change of time range. So what we've done is we've created two-way binding to where it feels

16:35 like it's the same app. And people might not even know that it's spot bar you know with mods and stuff. And this is the JavaScript API, I'm assuming? Yeah, you use the JavaScript API. Yeah,

16:45 that's what enables that. And then we kind of did something fancy, like kind of created a real-time interaction API on our product side. And so now you can build something that feels like it's part

16:56 of Wyshrock, but it's attached in real-time to the database. So if you, you know, imagine you're trying to say, like, let's look at, you know, uplift from, you know, well work, right? And

17:07 you might have a scatter plot of like how long, you know, how much production was lost before you, you increased it and you see this outlier. Well, you click on that outlier in, you know, in

17:16 Spotfire, which who knows how many ways you'd want to look in, you know, maps or whatever in that data, we don't want to build that. Well, you click on that dot and you can, and then Wiserock

17:24 flips immediately to that well in that time range where that thing happened. You're like, oh, wait, the capacity curve after was too steep or like actually that was categorizing correctly. You

17:35 fix it in the common source of the truth you click refresh and spot fire, and immediately that dot moves down. And so by doing the analytics, you get the human and the loop feedback. You can keep

17:46 building more and more advanced data science models, et cetera. And I mean, with all the amazing things that you can do in spot fire so fast, but you get to store it in a modern place with all the

17:58 interaction and talking in Wiserock. And so we really are like an amplifier for the biggest data science teams, where they're building so many models, but a hard part is how do you deploy it to

18:10 your users? So deployment and maintenance. Yeah, well, and I mean they can, it's pretty easy to maintain scripts, you know, and Python and models, but to get it to an interface where people

18:19 can see it alongside our data is harder. So we can expose all those model results as trends or as, you know, ribbons. And we,

18:31 and we also have an enterprise license so that it's unlimited users. And so it doesn't matter. some companies have over five hundred users and so that now puts it in the hands where the data

18:42 scientists like Oh Dang my models getting Used and I can myself build a user interface that feels like it's part of that dang now I don't have to spend as much time on Stuff I didn't want to spend Ram

18:53 on data monitoring is cetera

18:57 so that's been that's been that's been fun like Ransom you know we don't we're not we're not slowed down if we try to build southern reacts like every other software company does only then we would we

19:08 would not be able to iterate quickly and figure outright what is really good before we pull it into our core app if it if it becomes clear that it should be interesting it's cool the ER why assume

19:22 this integration also is one of the ways that allows you to share those exact views and all that stuff with when they have the communication to write like it's one thing to share a link to a spotfire

19:36 template But it's another thing to share the exact link with the exact filtering and everything, exactly what the other person is looking like, and bringing them right back to that state. That's

19:46 what I'm saying. Yeah, yeah. So I mean, even Tipko, when they saw us fall fire, now they're back to just spa fire. They were like, we never saw anybody. We never saw any of our software

19:57 vendors do that. That's awesome, you know? And so, but it's, the reason we did it is because we don't want to lose the ability to rapidly build stuff So that's, so that SpaFire aspect integrated

20:07 with Wiserock is that y'all are managing, like hosting SpaFire? Yeah, so it's our SpaFire server. You know, we are multi-tenant. We are, yeah, now they're Linux. So we're not multi-tenant,

20:19 every, you know, the database, every single operator has their own virtual private cloud, including their own SpaFire servers. We stand those up programmatically now. And we manage them, we can

20:29 beef them up however we want to And so

20:34 we pay, you know, a per. user fee to Tibco, based on monthly active users if somebody clicks and looks at our watch, at our dashboard and watch rock that month. And we had to make sure that our

20:46 user count is recorded and we pay based on that. But I mean, the value is dramatically higher than the cost. So I mean, anybody building an analytics product, man, it's a pretty good idea to

20:58 start with Spotfire to figure out what you want. Before you, you probably should and need to build an actual real software platform. Mess and various buys you first. Yeah, that might be good.

21:12 But that's totally possible. 'Cause it has happened. It has happened. It worked on that and they're awesome. So

21:21 that's been the spot for our journey. Okay, so there was a lot to unpack there. And you know, if we talked about wanting to get in some of the tech stuff, so we'll start you, 'cause you already

21:29 started talking about how you did a POC with D3, but you know, you didn't talk as much about it as you want, What did you end up on as far as the visualization aspect of the core web app and what

21:40 have been some of the benefits? Yeah, so one part that there's kind of a design philosophy that was developed over time, I also read a lot about the science of data visualization. So there's

21:55 actually, I didn't know this, but there's actually a right way and a wrong way to transform data in a database into the preattentive attributes, the colors, the shapes, the different types of

22:08 graphs so that it can go into the human brain as quickly as possible. And then, and if you do it correctly, the human will comprehend that data 100 times faster than if you do it incorrectly. And

22:22 so, man, we push data is as far as possible because what we're trying to build first is not actually an AI system you're trying to build artificial intelligence. What I say is that we're building

22:35 and I a system which is an intelligence amplification system where you're assuming the human is in the driver's seat and so it's not Irobot zero bottom there is man like you use watch out for the

22:46 first and you're ten times more capable than you or the day before as a human and so it's interesting that's all about data viz if you're trying to get data into a human brain by far the highest

22:57 bandwidth one is dative is just not going to be audible it's not going to be touch otherwise which can fuel tables Yeah Yeah and so it's definitely not tabular Yeah and so so then okay great now the

23:07 human has the insight but now they have to act on it let me as any to get something back into the machine and there's a there's a date of his principle called direct manipulation that's where you

23:17 would like like click a dot on a scatterplot drag it or click on a curve and move it up or you know last of something and and most software is built with forms where you'd have to type in those

23:28 parameters whenever on but that slower right and and so we've There's also some really interesting things about the way the human brain works that latency is really really bad so if you can keep an

23:40 interaction down to around three hundred milliseconds then it feels instant if it goes beyond more than two seconds and it actually gets it's where the human can't keep what they were looking at and

23:51 working memory so if you're flipping between walls and you have more than two second loading time between wells then your brain is not remembering the prior well and and I mean even better be

24:04 trellising and things like that but but what's interesting is and so we we basically found ways to get rid of loading indicators even one customer we just ingested and I think it was a yeah ten eleven

24:20 terabytes of skater and and using the technology that we have into it's compressed down to three per cent of that size but we actually can navigate that In NC you can see ten years of skate and wise

24:34 rock across twenty trends and it shows up on your screen and less than less than one second and so for the first time you can use your skater in an exploratory way and that's all possible but guess

24:47 what it's not possible a snowflake databricks those those tools are built for large batch jobs and we actually are moving that as part of our stock for the batch jobs that you do a few times a day

24:58 when it comes to real time interaction if you want to flip between you know one thing you do is put a memory and spotfire you flip between a very fast once you get big enough data it won't fit in

25:07 memory on your machine when all you need a database that can that can return data that fast and that's where we're using a technology called Timescale DB and that was because of our design needs we

25:19 have to remove the latency otherwise the humans won't stay in a state of flow the engineers won't be able to enjoy their job as much then we needed to find a better technology for that and so so we've

25:30 We found that by building an AI a system without any special algorithms or AI that man people just use wise rock and they love it they just they it just feels good it's almost like fun to go in and

25:46 and clean up your data I mean I have the the same defect from like I just wanted clean something like he had one is your say it's weird I actually was using wiser on a weekend Arctic cathartic you

25:56 know and so so that that design principle as actually made US build very differently like an AI system that's mostly trying to keep the human out it's going to surface stuff unlike a card you look at

26:09 Google's material design system there it's built for like that thinking and so there's tons of white Spacers I mean Davis has never used correctly and and you would you would start with a completely

26:22 different everything or here if you're saying I want to build an AI system and so Yeah that's the that's the background the database we chose like I said as we do everything we chose choose everything

26:33 open source except for spotfire so our databases post cross and with the time sealed with timescale DB as an open source extension on topic can you speak just real high level and timescale women I

26:45 know generally yeah for for people may not know Yeah so what timescale does is just say that okay Post gress is really built for transactional database that is with a real time read write and then

26:57 you've got things like snowflake and

27:01 you know databricks which they are optimized for large analytic queries so you have you're trying to save some i production across ten thousand walls and ten years of history in post Gress you know

27:13 even if you try to optimize as much as possible little job I take thirty seconds at least and if you did it in snowflake it would take probably less than a second and so what they're doing is OK how

27:25 do we take the best of both worlds as an open source Framework that that moves from a row aren't at storage to a table that is columnar and it'll last also lost that compression to where you can take

27:39 literally billions and billions of records of skater and compress it down ninety seven percent which means there's less read write on the on the disk and it's actually easier to uncompress it than it

27:49 is to read from the disk and then the just the query patterns work better and so you actually get the the best of both worlds you can built you have to build a sauce up on a rewrite database because

28:01 you have to write back and writing back to snowflake very very long time like I mean thirty seconds type thing and so when we get that and then in the same database so we can write one SQL query you

28:13 could join the thing you just now wrote back read by like dragging a variance event with as much scared as you want and all your while view history and that comment there's a I extensions on top of

28:25 post crescent so If you're interested in databases, Google Postgres for everything, there's really, I think, a very strong case that you don't need to build anything other than on top of Postgres

28:37 because it's the most advanced open source database. And there's tooling that can do memory caching and vector databases and - You know, you have a special analysis. Now I have a DuckDB extension,

28:48 so now I get some of your OLAP type stuff that you're talking about. It can be done faster. So, I mean, it's way better to have one tool to maintain than a bunch of different ones, and so - Or

28:58 having to replicate out of one to do that job? Yeah, yeah, and so that's the back end, and then we use DBT to manage all the transforms as code, which if you're building a data warehouse or a

29:09 data tool, like there's kind of the default, there is no other like close, you know, technology than that. And then we use DAGS or for orchestration and other open source tool. And then for the

29:21 front end, we use React to build our interactive visualizations. We we need pixel level control and to be able to to build that direct manipulation be able to drag elements and keep it real time to

29:35 where you know as you're dragging the other elements on the screen or updating as well and do all that is possible you know with modern you know architectural Amp like a react so it was going to move

29:48 up his act you're talking there I mean which we've talked about DVT like rob over and I've talked about it because we use it at Chris mill and there's some new players in the game and if you looked at

29:57 SQL mesh or any of those at all yet but I've seen a little bit but at the same time it's like once you already implemented a DVT you're not going to get the benefit necessarily but DVDS been great for

30:05 US except all the cloud offerings are getting more expensive yeah but the DVT Core Yeah it's a python it's free Yeah Yeah but Subodh Daxter cause that has not come up think maybe we talked about

30:18 Airflow a little bit but men didn't get what was earned me an evaluation there and why he chose daxter it and if you looked at airflow also or semis others competitors out there. They've been

30:27 marketing because I've been getting it. Or why not just these kind of ads? I mean, we just have a theory that actually open source is gonna win because people, I mean, a lot of developers are

30:36 very cost conscious. Yeah. We aren't, if it's worth it for another company to pay for this tool, then it probably is for us too. So don't account for the price of the tool, just pick the best

30:48 one. 'Cause for a startup, the biggest cost is cost of delay. If you could have got a feature out that's worth a million dollars a year, three months earlier. Yeah, I know how you money. That's

31:01 very expensive to make it wait longer, way more than whatever you're paying for that sort of thing. Or did you lose it on a customer? 'Cause it didn't exist. Yeah, or you miss the turning point

31:10 of getting traction in your industry and somebody else beat you. But it's more that, hey, let's try to go with the open source. and and that's so that's why we picked daxter and it's worked really

31:22 well explain what dexter is for people who may have been but in that case a airflows open -source to write so like say dogs are a little newer and I I don't have as much knowledge about the the

31:34 orchestration differences but but effectively like you know DVT is like okay here's your SQL scripts that when you say go I'm going to transform the data this way orchestration tells you when to say

31:44 go like it detects that you know our process that that pushes to US from our customers dropped in an s three bucket or in the snowflake if they're using snowflake and so now it's going to trigger a

31:55 DVT job and I mean a a daxter job and I don't think it's as exciting as the as the DBC side but it is a really important but my DOCs are much like manga DVT is built on the concept of dags right Yeah

32:09 he erected a cyclic graph whatever that just been because an order basically but Yeah and then but it's like there's so many things that are based on that react reactors based on dogs XXL is based on

32:19 the gas and and so is you know orchestration you know it it all should be all the item thought of as a dagger and directed a cyclic graph but what what is interesting is man that the space is changing

32:33 so rapidly that you know what I really think is man and oil and gas operator or really any fortune five hundred company man why would you want to build your own and like build like what we're not

32:46 building the data warehouse for a whole operator but like the core data will we we've already built the tooling that goes to all the systems you probably want to pull from and then even if they built

32:57 something a quick or excel I mean we can build a connector that you know in a day and now it's in but you know what happens whenever doctors not the right choice or we should move to sequel mashable

33:07 because we manage everything is code we can rip that out yet of the floral industry and and then put something else in transparent it's better or Yeah the customers don't care they don't know like

33:18 we're the users of our data stock our data engineers Yeah and so if we see something that's going to make them more efficient and more effective then we'll switch it out and we can't because we built

33:28 it to be able to do that and and they're they're developers right and so if you if you aren't building it with developers and you're using any like you know graphical user interface to do transforms

33:39 then they can't do that you can't factor would be you can't parse data and push it into a different source of meta data like whereas you can generally use other tools M and M Oh Yeah I didn't mention

33:52 that we do owe the the cool thing is that where we're actually up to about that and that it's it's fifteen percent of U S onshore oil and gas is using wise Rock Nowa to manage their production that's

34:08 Awesome and every operator is on one hundred percent of their wells and then we have expansions some of our biggest customers keep getting acquired and then they're expanding to the companies that

34:19 acquired them a couple super majors and we see the same ability to build these data warehouses as code crazy fast and performance doesn't degrade and so you can imagine we think we could get to 50

34:36 market penetration in two or three years.

34:41 Well if we have that core component of a data warehouse then and we have the ability to manage it for you and you've got to if you're an operator with a data engineering and data science team what we

34:53 try to do is make it towards very modular so you can actually use a data mesh type architecture and they can have their own dbt stack or on snowflake whatever they want and be able to pull in the

35:03 metadata from all our transforms and have the work administration you know work between them to where it feels like everything we've built is is just a component of their. large data warehouse

35:15 ecosystem. But they don't have to build the core stuff that's common to every operator. They get to work on the secret sauce, just stuff that they think is a competitive advantage. And so that it

35:25 does seems like it's actually a business risk to say like, no, let's build the same thing that everybody needs ourselves and then try to stay cutting edge.

35:36 Because I mean, now with AI, it's changing even faster. Yeah, sure.

35:39 So I mean, there's been two major seismic shifts in the modern data stack just in the last 12 years. And I think that's gonna happen roughly at the same pace. Where you probably do need to rip out

35:50 large parts of the stack and put something else in, because it's five to 10 times better than the prior version. I mean,

35:57 like you said, data warehousing and comparing snowflake or data bricks to tear data to things before it. Or even to register between that. Well, and there's arguments that you don't even need

36:07 snowflake for a lot of stuff. You know, they call it big data's dead and that's the DuckDape DB thing. And that's a Postgres thing because now. When you can spin up a a cloud machine that has two

36:16 hundred fifty six CPU cores and ten terabytes of memory or something like that and that did not exist whenever you alright someday invented redshift so they had to build horizontally scaling system

36:28 yep but that actually creates a lot of headache and yang reduces all the communities under the hood knowledge of why not just scale up why don't just pick a really big machine burglaries and and then

36:37 you look at how big that is almost no company actually has data bigger than that now even the supermajors woman and why go to the complexity because there were that article from I think it was Georgia

36:48 George fraser from five trend but put it out there there was a whole big data is dead but mean that they had all the queer history from what snowflake and red shifter and it was laugh they are like

36:56 ninety miles or even ninety nine per center of at least ninety plus percent of the queries were less than I mean it was it was so tight I mean because most worries are not that big you see all the TPC

37:08 Benchmarks and on one terabyte or ten terabytes of data like no one's really doing yeah he may know him yup no one in that hole Yup take the parade of principle most people are not even touching that

37:18 close and again you can run Duck DB on your laptop and get similar performance or Yeah so we're actually self -hosted we're we're not using RDS for progress we're we're spinning up easy to emphasise

37:28 that are reserved which eight of Us let you reduce the cost by like sixty percent Yeah you have to stay on the same size easy to instance you just commit to like a spend and then you know order were

37:38 you know obviously managing the post cross at all that with

37:45 the the infrastructure as code type frameworks and so we can you know we can add a a CPU core to a post -course instance for like twenty bucks a month and in and that's just insane like how our costs

37:57 are so so low and yet where we can ingest ten terabytes of all the skater from a super major for every single one of their Permian sensors right and then the performance is sub second no matter what

38:12 you want to do and so I would say like man why And we get the read right, you know, we don't have to worry about the constraints of a snowflake. And so I think that's like the type of example where

38:23 if you realize that happened, it's like, okay, first there was non-cloud with terabata data, then they had to invent redshift, which led to snowflake and Databricks. But now there's people

38:33 arguing that that is the wrong way, go back to a single machine. And I think they're correct. And that five years from now, that will be the trend. And snowflake and Databricks, they're moving

38:44 to that trend too, they're gonna figure that out. But

38:47 okay, what's gonna be the next shift? It'll probably will be something to do with AI and metadata, things like that. But no one can predict it. Everyone's gonna add vectors to their database.

38:56 Microsoft's already starting to do that. But again, I think that's, but again, it's, to in my opinion, the incumbent databases, the Postgres, the sequels, there's no reason for them not to

39:04 just have these extensions instead of, oh, well, like for us, I've gotta use quadrant for a vector database because it's optimized for that. right and it's the fastest one that we have found but

39:17 there's no reason like we looked at PG factor at the time it was slow or but there's no reason that Postcard doesn't buy and then enter like that will happen or could happen at some point like an open

39:29 -source framework that you know they are they are improving crazy fast because there they have companies on top of you know timescale as a for profit company but everything they're building is open

39:39 source as well because most companies don't want to deal with the doing it on their own a piece about the Injustice I mean so we were messing with timescale and RDS I were very infancy of times guilty

39:51 but we didn't think about them I didn't have a knowledge of the time that our developers the contractors didn't have the knowledge to deploy it you have like much like you guys know now but but I mean

40:00 I think the benefits of it and it is a time series database so we didn't really get into when you're comparing that said Influx DB years and these other things were those were no sequel but me this

40:09 one like what may cause it doesn't

40:13 Uh,

40:15 it's like hyper something, hyper tables, hyper tables. But like, but the ingest is a huge deal for a time, for a time series database. Cause it's, you know, it should be collecting all that

40:24 IOT data or skater or whatever else. So again, it's optimized for getting in where, like you said, getting large amounts of data into snowflake has to be usually these batchy type things. And

40:33 they're getting a little better now with like snowpipe streaming. But again, those are still, you're paying for ingest. Um, and again, if, if you can get similar or just much better performance,

40:43 and you're just paying the fixed cost either way, um, yeah, yeah, yeah. And so, you know, I can, you know, that's kind of, um, you know, that I can nerd out about that stuff all the time

40:57 and I don't, I don't write any coding more, but, you know, I think that, um, you know, for anybody listening, and this is obviously a, a nerdy podcast. So probably technical people were

41:07 still listening right now. Um, but the, uh, the thing I would. I would say is, you know, the thing, I think that was really important for Wise Rock is, you know, choosing a domain that like,

41:21 you really want to know enough about to be competent, you know, to be an A plus player, you know, it's a minus production reservoir engineering. And then, you know, a little bit of competency

41:30 and like modern software engineering, again, like probably B minus and that. And then, you know, data is, is very different than software engineering Yeah, so like, but you combine those three,

41:44 turns out there's a lightning rod of innovation where you can take the best of all those. And really, combined with modern data stack is kind of a separate part from the analytics. And that's where,

41:55 you know, we've been able to build this product that is really so simple. We don't have any direct competitors doing that. Like the people in the production optimization space are building

42:07 algorithms that will tell you like they optimize the CSP, et cetera. and just that data too, and show it. And now it's all in one place. But there's no one else that's saying, okay, we'll just

42:19 pull it all in so it's easy and fast to see. And it turns out almost all the time you're just looking at data, you're not actually needing to write back, except for a few key things. And so, 90

42:31 of the time, if you need to know what's involved with you, just look at WiseRock. By exception, maybe you need to go update something in the well view So,

42:39 it's been

42:42 interesting how the legacy of the Spotfire development has really influenced the direction of the product and that it's been received very well. And even when people see it - you know, super majors,

42:59 if you show an algorithm, their data science team will say, I can develop that algorithm, too. Sure. By the way, you do a trial. They can definitely build that algorithm it's not that hard to

43:07 write a Python script and ask chat GPT how to do it. But if you show But when we show the front end, man, that requires real software engineering, structure, team, and for structures, no one

43:20 says like - There's one EMP in the world that I could think of that maybe has people. Yeah, yeah. And so that's been - It would be willing to invest in that side of it. Right, right. So that's

43:32 been a

43:34 good thing. And the value is really there

43:39 And the first year and a half of using Wiserock, they increased their production by 2. And

43:46 that was without AI. And the reason they said that is because the

43:52 VP of production there, at the time, Shadd for Azure. Shadd for Shadd. Yeah, he's awesome. They would have a weekly meeting before Wiserock. And they would have kind of a list of walls that

44:05 were down according to a rolling average. He'd say, Hey, why's this wall down? And you better know, right?

44:12 of all, it's a way better target than a rolling average, but there's a simple dashboard that was built that it would be read if there was no comment about why the wall was down. And if there was a

44:23 comment, that table would have the comment. And so, and it would be show the engineer's name next to it. And they said, Hey, like before the meeting, all your walls with more than 10 barrels of

44:32 production losses need to not be read. And then, and so then like, I mean, shadow come with that meeting, and they're going to talk about the fifth wall on the list is only one actionable. And

44:42 then the 13th, right, and they found that the remedying, you know, wells that need to work over is went from, I think, like 16 days down to seven. And that was just because of the communication,

44:55 right functionality. And so there's so much you can do with just removing friction of human to human communication and machine to human communication without having to worry as much about the AI But

45:09 we do plan on the cool things we have the foundation to be able to. to start doing some of that stuff too. 'Cause now you have all the context per well, you know, and whether it's tagging the time

45:19 series data, but then throwing on top of the conversations and everything. Yeah, no, that's, I mean, that's the thing, right? Like that's why where I get bullish with the AI stuff is it's, if

45:28 you have data with context, that's exactly what these things want. They want all the context, right? And so like, and yeah, like you mentioned, 80 of the AI stuff is cleaning the damn data and

45:40 getting it structured and doing all your ETL and all that stuff. And so I feel like as an industry, we

45:47 holistically, the industry has kind of pushed back or been disappointed with AIML promise so far. But I think that's mostly due to us because we all had the shitty data and they're like, oh, let's

46:01 try this out. And then they did it with bad data and they started down that path and we're like, oh, we have all this work we need to do to the data before we can ever get to the AI stuff. I think

46:13 we had to learn that the hard way, unfortunately, before we really started. And I feel like now we're kind of coming out of that trough of despair into like actual real world applicability and

46:23 useful functions and use cases and things like that. But ultimately it all boiled back down too. We have to have good data, like it. Yeah, yeah. Well, I mean, if you, like one question we

46:34 usually ask an operator is, hey, do you have the history of artificial lift types and equipment in each well over time? 'Cause we can display how the bottom is a ribbon. It'll, you know,

46:44 underneath all the production you have this little, you know, horizontal bar that's like, during this time you had an ESP and you hover it over, it tells the, you know, whatever metadata you

46:53 wanna attach to that is there. The problem is we say, you know, are you storing that in well view? Like, I think it's been about 15 of operators that said, yes, it's in a database. And we said,

47:06 well, that's okay, is it in a spreadsheet? Maybe that adds another 20.

47:10 the vast majority operators have nowhere other than, you'd have to go through a bunch of emails of like what even like artificial lift type was on the well, much less like the really detailed

47:21 configurations. And so yeah, how are you going to build an AI model to look back and try to optimize like when even, of course, that means lots of other really, really foundational data isn't

47:35 there, right? I mean, like you can only determine so much from pressure trends, like even major events that caused, you know, a million dollars of deferred production across six walls right next

47:46 to each other will have zero piece of metadata that even talks about it for one sentence. Right, we got like a crack here. The vast majority of companies, right? Yeah, like the frack hits are,

47:55 you know, you can determine those, but like, I mean, somebody was talking about it in email, but like, that's gone, right? And so with Weijerock, we're getting that back in. And once you're

47:60 there, then it's fairly easy for the AI to.

48:09 you know something we're we're looking at is like okay you have a variance event go look through all the data next to it can you can you didn't ask what it is oftentimes know what we have that little

48:19 chat functionality the AI could just say oh there six people that might know about this well go Hagdom Hey what's going on and then they can reply in their email like OH it's this boom EH okay Cool

48:31 I'm going to categorize it as this now you don't reply to this email if it's correct if it's wrong Lemme know right and so like you can imagine how how straightforward that would be where the law they

48:42 don't even need to open why should I look at their phone as matter where they are you know in the field or whatever and and they they will be adding the context unnecessary for you know everything the

48:54 AI needs to do to get better Yeah and now you can start to do the analytics of My Hey long how long does it take Us to you know do you repair any SP and get a new one in the whole compared to everyone

49:06 else in the Permian right because we actually know that This is how long it takes, whereas people can't ask those types of questions. Even something that basic right now. Yeah, and then I mean,

49:16 like one thing I want to ask you, like are you all bringing in anything around, like say, LOE, like even like cost and stuff like that yet? Or, I mean, didn't be able to back in the napkin

49:24 economics for certain things or - We're not doing anything with LOE yet. You know, that's on the roadmap. You

49:30 know, the interesting thing, you know, that the hard part of Wyshrock is, you know, scaling the company. Yeah, you get in focus Yeah, and so we've, like, we've really been more constrained

49:41 by, you know, learning how to hire amazing people and then how do you work together as a team.

49:48 And what we found is that product market that has not been the problem for a long time.

49:55 And what we're not gonna do is just scale as fast as we can by hiring a bunch of sales guys and then see the culture of the company like completely degrade or the product quality to completely degrade.

50:09 We actually

50:12 have not had a full-time salesperson for a very long time, and it's grown through word of mouth. And what we found is that, okay, we shouldn't like make it harder to understand the value of Wise

50:25 Rock, 'cause it's pretty darn simple right now. And really, the more features you add, the more people in the room need to say, like, yes, I would like that. So we found a counterintuitively

50:35 like, man, if we just really laser focus on this one thing, well, then you can get that market penetration, which is more important that you get their relationships with executives. And by the

50:45 way, the executives are using the results of Yshrock and they understand it too. Like the CEO, Lance Robertson, an endeavor, man, he loves Yshrock and he, the former CEO of endeavor. And he's

50:58 like, man, y'all helped us transform the culture of our company. When you get that type of relationship and you come back later and say, hey, we're gonna go and now to the LOE. side We're going

51:07 to start getting some of the reservoir data in here, or, um, we supply chain, whatever, um, well, then that's a lot easier with the, they trust that, you know, you're going to do an

51:16 excellent job with that as well. No, for sure. Yeah. I just, you know, curious. So then maybe people taking the data out of wise rock to do say analytics, cause I mean, when you're talking

51:24 about how long is it taking us to do the ESP? Cause now like funnels into like field development, planning, and then like setting budgets for the year, like, and just, there's all this stuff,

51:33 like kind of downtime assumptions and everything gets baked in So I'm assuming people are using those outputs to enhance those processes also, so. The standard way that a SaaS company would give an

51:44 operator access to their own data, if they do, is to charge additional for it, and then do like a daily dump to somewhere, right? Well, what we do, I mean, you can architect all this in a

51:56 secure way, is we just give them a direct SQL connection to the live database. We could also even do a read replica needed, but that's not even a problem us right now. And so.

52:09 It's actually real time that if any of their systems, someone adjusts the capacity curve and their system wants to know that happened, then it would be immediate and they can pull it again. And so,

52:20 and it's all in one database. So a single SQL query can join all that data. And I think we have, it's definitely less than 20 tables that store the interesting data in Watch Rock. And I mean, the

52:32 data model is just very, very simple

52:36 When you're talking about these core data sets, and so it's

52:40 an easy one to integrate back into your systems. Nice, we've already gone through 55 minutes. Yeah, no, no, it's so crazy. The only thing I wanna hit before we get to the speed round and you

52:52 thought of it earlier, I'm glad I still remembered. Some people may actually know you're behind it, but you can talk a little bit about the Iron Python reference, the website. Oh yeah, yeah, no,

53:03 that was fun So starting wise, Rock.

53:07 You know, like I said, I found that if you did, you know, Iron Python lets you automate everything you can use with the graphical user interface in Watch Rock, everything you can do, you could

53:17 also just do with code. And so it allows you to build these data products that, you know, rather than, you know, building 10 tabs, why not have a control that you just slice and dice the data

53:28 right there.

53:30 And so Iron Python was like really critical. And I, and so, you know, I was like, man, there should just be a reference that for every user interface screen, it should show the snippet of code

53:42 that would let you do that same thing and that didn't exist. And so it's like, oh, this will be a great way to get customers as a spot for consultant to that was a bad idea, because anybody who

53:52 cares about Iron Python, it's not the one hiring a spot for. There you are, the spot for consultant and they don't want to hire another spot for consultant. And so, but it was fun to build and,

54:03 you know, But unfortunately, has not been completed. one probably somewhat obsolete as versions of another spin enhancements think by the Yeah but nonetheless like because Yeah but I think at the

54:14 time that you did it to spotfire documentation and the community sites were very dismal or not great you know they've definitely come a long way Yeah Yeah now Chachi P T I haven't tried it but you

54:24 know it could probably do a pretty darn good and may have to take a screenshot of the Gooey draw an arrow to hey how do I change this one with a spotfire and Python expression I bet it'd be pretty

54:35 well may have their own copilot too I think Java or go pilots of MIa the world changed a Latin were a private eight ten years they wrote that but I was definitely pretty cool like cause i Gotta think

54:45 about the time I realized you and I were playing in the same sandbox like Oh it broke actually this one who who the bill arrives a while and one thing I wanted to say before the into is we're men

54:58 we're really focused on just building an awesome culture and awesome team at wise rock and what we're trying to do is is hire the very best Patrol engineers and if they got you know more the digital

55:10 experience all the better the best software engineers and the best data engineers and bring them all together you know and solve problems that can be deployed right now across fifteen percent of U s

55:21 onshore oil production so if you're listened to podcasts and this was exciting mean we love to we love to talk to you and reach out

55:29 to US our art drugs for people offices is Kate Steinmetz Okayed at wise rock software dot com would love to talk to y'all and of course if you're an operator love to talk to you always do but awesome

55:43 man now we we typically finish it with just a speed round of kind of silly questions for you to answer just to go and buy silly I mean still Nerdy tech related ones but they're just awesome food or

55:56 Yeah but now alas since we didn't get that you mentioned a ws what what's your favorite or preferred cloud provider I mean we chose eight of US because it had I think seventy percent market share at

56:09 the time and I dunno is price still the case because they're all growing but like wasn't there they were bigger than the other two or three next combined as far as compute that Yeah I Will Admit I you

56:18 know Microsoft and you know Google Cloud and data it was just far bigger and Nina and we weren't going to be on the you know Microsoft stocks are wasn't really a reason to choose azure and Google was

56:32 only like ten percent of the market Yeah probably I dunno that binge Batman they just they rebel features and is just don't Google unless it's like a very stable product lights or something ever

56:45 became clouded pigging out of the US or they're all really good at this point Yeah but I was your favorite Open Source Library

56:56 and

56:60 I think honestly the the DVT Side is I I Haven't gotten to write a lot in it myself but just what it Yeah what enables is crazy and it's really fun they they're basically bridging the gap between the

57:13 bases said hey we're a bunch of you know analytics developers who knew how to write SQL but then there was a few of us that had the software engineering side to realize that if we combined software

57:24 engineering best practices with what we're trying to do a sequel then we could build something crazy it's kind of that lightning rod of innovation as somebody somebody set across both of those going

57:35 to say there's the Iran Man it's a beautiful thing and it it's changing it actually has changed the whole modern data stack has to change to adapt to what DVT is making possible so that's a really

57:48 cool thing and I like that there's not really a cause close competitors with it you know everyone you know can just use that and there's not choice paradox of

57:58 OH and what was what did Y'all do on on the snow day this week with your shop Yeah My Do We We we drove about three miles and there's a there's a big hill you know north of north of I tend just

58:17 outside the beltway and we had this little

58:21 blow up a two person boat on the beach and then a boogie board is about two and a half feet long and so we start at the top of the hill and they they just went down over and over it lost it it was a

58:34 ton of fun for about twenty five minutes but we had rubber boots and pairs of socks and two of them got snow in there and they were crying and I don't blame them now so it took them all to thaw out

58:47 and then they had a lot more fun we actually have a for roughly a two foot diameter snowballs in our backyard right now I think I I think it's going to take about two weeks to melt of because they're

58:59 so big so that was you know fun stuff like that Yeah that was that was Art favorite thing my daughter got to build a snowman for the first time. So it was, it was pretty cool to, to go do that.

59:09 Houston, I had never in my life expected to use them to have that much snow. And so it was, the only thing that's more astonishing is Louisiana. And my mom isn't Florida. She sent me pictures at

59:20 Orange Beach and like Gulf Shores got like eight inches. My brother is crazy. And near Tippito, 10, 12 inches. Like, yeah, there was a blizzard warning in the south of Louisiana, which is

59:31 crazy You know, all right, I guess last one, what's your favorite restaurant in Houston? That's a good one. Um, you know, I, I really like a folk at a chow, just because, you know, all the,

59:45 I really like variety. It's all to me. You could want all the variety. My wife doesn't like it as much because she's like, there's no way I'll get my money's worth enough, but, uh, but that's a

59:55 good one. Um, and, uh, yeah, that'd be my first answer that comes to mind That's a good one. where where can people find you if they want to reach out or get in touch Yeah do you know Linkedin

1:00:09 or you know Brock at Weiss Rock Software Dot Com is a Wise Rock Dot com Wise Rock Soft Dot com Yep Yep and I and We're We're located at Bunker Hill an item we love to have you over to our office

1:00:23 anybody can write down the road to visit so Yeah Yeah and then also on the data science software side if they're interested Yeah no we loved it we loved to to and and and meet you also meet meet

1:00:37 anybody that's interested in exploring that type of thing you know I think I you know it's just it's anybody that that has the oil and gas domain knowledge and you know even an interest in in

1:00:50 analytics and you know learning something like SQL I just think the Sky's the limit and if you can find an operator that would employ you to actually do that which unfortunately that's most operators

1:01:02 would not Like be that interested in the Potomac near learning that stuff but there are they are out there and man I just think and you know go for it and and do you have an opportunity to get into

1:01:13 consulting side of things even more just go for it and the the downside is is pretty limited in reality and and so you know we we have a team of you know a bunch of digital Pajama chairs that would

1:01:27 love to to talk about that too and in their experience they're so beautiful that's Awesome that's what I love to see Yeah all right everybody thank you for for tuning in Google or Youtube thinks

1:01:38 you'll like one of these videos around here please like and subscribe we've seen a bunch of growth so we appreciate you guys and remember to do that we will see on next time