Subscribe
Share
Share
Embed
Welcome to Energy Bytes with John Kalfayan and Bobby Neelon. Your essential guide to the intersection of data and energy. This podcast dives deep into the world of energy, shedding light on how data, AI, and technology are revolutionizing this sector. Each episode equips listeners with insights into the most efficient tools and resources, paving the way for a data-driven future in energy. From technical nuances to broader industry trends, Energy Bytes offers an unparalleled perspective on the evolution of the energy industry. Join us as we decode the algorithms of energy, one byte at a time.
00;00;01;21 - 00;00;26;22
Unknown
Wake up. Welcome. You can start with any kind of current event. Things that we got from AWS that got to be out there, Duck Labs. I know. Prefect bot Daxter. I didn't have that on my bingo card. I didn't have a lot of it. I mean, even I don't even think we've talked about the grok, cursor stuff either.
00;00;26;25 - 00;00;45;25
Unknown
Oh, yeah. In a while. And just like how big brain ends up looking. It's like, okay, we'll just control the whole fucking chain. We'll go from hardware all the way up to actual software. People actually writing this over. Yeah, well, like, I and I downloaded crockpot today. Just like, fuck it. Let's see what what this is about.
00;00;45;29 - 00;01;13;19
Unknown
But it's of course, only running locally. And so but I think that's like, there's such a everyone I talked to you in like at the office is like, yeah, if we could figure out how to easily put this in a VM. And so I could run my clock remotely so I can always be running my clock. And so having to remote control from my laptop and then yeah, when I'm like it's annoying at this point I've got a hotspot that I use for clients and I'll turn it on in my car and let my laptop keep working.
00;01;13;20 - 00;01;32;24
Unknown
Yeah, yeah. Like I'll commute home because I'm in the middle of running something, right? Yeah, this is stupid. You can just do this in a VM. It is why I'm. I mean, with just the other day, I realize with cloud code that I get running, and then it's like you can log in to your cloud app now and, like, see, everything is happening, and I'm running Codex right here on my phone.
00;01;32;26 - 00;02;00;03
Unknown
Easy. Oh, I'll have one laptop benchmarking local or local models on it. And that's my Nvidia gaming laptop. And while it's doing that, I can remotely control it from my cloud desktop app on my MacBook, while I'm also got for tmux or c mux windows open running and Codex also running. So it's just crazy, man. But that's where everything I feel like is going.
00;02;00;04 - 00;02;23;00
Unknown
Like enemies. Added the bots right after bugs came out. Yeah, the whole agent working together thing. That's exactly what Grok Bot is. But they just give you a VM because they have all the fucking hardware in the world. So it's it's fascinating. There's lots of stories there. I don't know which which ones we want to, I mean, I feel like I get which one is most interesting to you.
00;02;23;05 - 00;02;47;27
Unknown
To me, it's the running VMs consistently while also doing other work. Because I remember we did that for, training our transformer models because it takes so freaking long. Yeah, that you have to. Well, that's like I was talking to our ML expert, and I was I sent her some stuff. I think it's called Hoodie Hoe dye. And it's essentially a genetic driven cloud infrastructure.
00;02;47;27 - 00;03;06;26
Unknown
So like, the agents can spin shit up on their own and yet without any approval of it, and it's, but then I was like, man. And so then I got her thinking of like, hey, we could actually use this to run parallel training instances or split up the training across like, topics and have them running in parallel.
00;03;06;28 - 00;03;26;10
Unknown
Yeah, all at the same time instead of having to crash everything on the one. Anyway, it's, Yeah, the whole agent thing. I think that's probably I mean, that's I think what we're doing right now, if I can just probably use the last five minutes. We'll all pause and do the intro, and you can tag it on the front of this.
00;03;26;10 - 00;03;53;23
Unknown
Jacob. Okay, everybody, we are back in the studio, for all together again. Yeah. For the first time in a minute, John Callahan here with with Bobby Nealon. What's going on, guys? Thanks for making the track. Especially with the, tropical whatever brewing off the off the coast of Louisiana right now. No, it's. Yeah, it's funny, because I, I guess, say potential client.
00;03;53;26 - 00;04;09;04
Unknown
You know, I'm going up to Oklahoma City after this, and I was he's in Oklahoma City. That's right. And I was like, I look at coffee like, actually, I'm gonna be in Houston on next week. It's like, well, I'm gonna miss you. But, you know, trip trip to Houston wouldn't be complete without a, a tropical, you know, storm of some type.
00;04;09;05 - 00;04;34;06
Unknown
Yeah, absolutely. Well, it is dry currently and not raining here in Houston, so hopefully we'll be all right. Yeah, we do need the rain. My sprinkler system would appreciate that. So any water bill. But, and actually, actually, you know, when we get into it with, actually a lot of a lot of things that are very pertinent to what may or may not come down with, with a tropical storm hitting so late, so perfect.
00;04;34;06 - 00;04;55;00
Unknown
Yep. Yeah. We're here today with Ariana Borelli. Yes, yes. Lovely to me. Thank you guys so much. Yeah. Thanks for thanks for joining us. And you're a founder or. Yes. You've already energy. Yes. Founder and CTO. All very cool. Well, let's just rip the band aid. What what is a Vardy and and what do you do for him?
00;04;55;04 - 00;05;25;20
Unknown
Of course. So my co-founder and I started a Vardy. And what we do is we work with utilities like PG, E, Kaiser, Ercot and, a lot of the co-ops in uni's in this industry. And what we help them do is automate a lot of the engineering that happens in the back end with regards to things like maintenance, things like outage management, planned outage approvals, and all of those things so that operators can spend more time in the field dealing with all the new load surges that are coming up.
00;05;25;20 - 00;05;45;14
Unknown
So much is changing in the industry. And so we want to help save them time and increase efficiency doing that awesome. Yeah, I mean, I, I think after especially even after barrel, I mean, people were just pissed at everyone in Texas. Like we we've had enough things happen like why are we not prepared for this? How can we not get things back on line faster?
00;05;45;14 - 00;06;15;04
Unknown
So exactly, exactly. No, I think that truly raised, a very or put a big magnifying glass on the, the whole utility infrastructure and the lack of investment in it over the decades and stuff like that. And so I'm glad to see because those are it's like it's such a hard thing because it, I feel like ultimately there's very little there's there's harder it's hard to incentivize a utility to do to spend money.
00;06;15;10 - 00;06;38;16
Unknown
It's hard to incentivize any business to spend money on things that they know might happen right. On potential outcomes instead of defined outcomes like, why would we have winterized everything rises until yeah, yeah, yeah, true. Like, I mean, we were fracking wells with UN winterized equipment up in Arkansas when I was in the field. And guess what? It sucked when it snowed the four times a year that it snows in Arkansas.
00;06;38;16 - 00;06;56;23
Unknown
Yeah. Because we'd everything would freeze up. Oh, and then it's like, you can't really do anything right. It's so interesting. And operator was speaking to mentioned that all of the maintenance is failure driven because they just don't have the time. And it's such a hard problem. You know, even on the operator side, as you're working with ancient infrastructure, you are dealing with.
00;06;56;23 - 00;07;20;16
Unknown
So much new low generation and demand, Ercot itself. I think the demand went up this year by 90 to 91GW, and it's predicted to go up to 365 by 2018. And then you have the low generation queues because of these data centers. And they're on 400GW separately. Right. So I think it's behind the meter. Yeah, exactly. That's a hard problem to solve on the operators.
00;07;20;16 - 00;07;48;23
Unknown
And as well, and so that's why I think there's so many smaller things that can be done in this industry. And we're here to tackle those. Yeah. No, I agree with you. Our industry historically, the energy industry is very retroactive. And so I mean, they're they've gotten more proactive with the basic things like maintenance and but like I don't feel like that's until the last, I don't know, 20 or so years we waited tell everyone got their general ax and all.
00;07;48;23 - 00;08;15;07
Unknown
But like it's I mean just across the board though, right? Whether it's, utility or an upstream operator, you know, they're not incentivized to spend money unless they have to rate on preventative things like that. And but ultimately, you know, over time, with all the other other industries out there. Yeah. Well, but like, you also have these, you know, like the manufacturing industry where it's like uptime is the metric, you do not want downtime.
00;08;15;07 - 00;08;34;04
Unknown
And yeah, it's like, okay, well the oilfield, it's exactly the same way utilities are the exact same way. Right. It's all about utilization. And so if you want but with little competition now too, it's like right. So what. Yeah. Like what. Right. Right. And so yeah it's such this weird balance you got power to that's all your brokers essentially.
00;08;34;05 - 00;08;51;16
Unknown
Right. Or whatever. I mean like that is what drives me nuts about it. It's all coming from the same. Oh yeah, I like the open market and it's like there's three companies. It's not true. That is true. Yeah. Like, who can I choose? I literally can't choose anywhere other than CenterPoint. Right? Right now there's very few areas where you have that overlap.
00;08;51;16 - 00;09;15;27
Unknown
But like you can choose the benefit here. 100% green electrons. They're coming through the same conduit. You know, my neighbor, he was hitting all the green electrons. Oh, right. And it's crazy because it's marketed a lot as a deregulated market, especially in Texas from the Texas area. It's really, really not. Yeah. No, it's not at all. If it was deregulated, my energy prices wouldn't have double their elastic.
00;09;16;00 - 00;09;39;02
Unknown
Yeah. Natural gas stays flat. It's like that's such a joke. Well, now, I mean, obviously we're getting into it, but like, the just the AI data centers is gonna add a whole nother beast to this and such big things. That. And then there's all the best systems that people are trying to throw up that, you know, are being marketed as more reliable energy for the local, you know, and businesses.
00;09;39;02 - 00;10;03;17
Unknown
And it's all American energy. And it's like, yeah, well, the energy on our grid is American energy. You know, like I don't understand that selling point, but it's, yeah, it's I mean, it's the Wild West is as far that's the I think the worrisome part of it is, you know, I'm all for making more uses of the electrons so that the grid itself can be more reliable and incentivizing things.
00;10;03;17 - 00;10;27;29
Unknown
But at the same time, the way that like Ercot is currently set up, especially with the on demand stuff, they're not set up to make the grid more reliable. They're set up to just capture the arbitrage opportunity when it goes from 50 kilowatt or $50 a kilowatt hour to 5000. And that's where the money is. That's not to help the local community get it.
00;10;28;03 - 00;10;52;02
Unknown
And that's not how it works. Right, exactly. And that's why it's also something we've noticed is harder for utilities themselves, because you actually won't believe this a lot of times with ISOs, they reject outage approvals and requests. The day before, and someone we were talking to, an Austin actually was mentioning on the generation side, they struggle so much because they have a couple of maintenance requests for these really niche, parts.
00;10;52;05 - 00;11;12;27
Unknown
And they're in wind. Okay. So it's more very generation heavy. If something breaks down, it is a really big cost as well on infrastructure and all of the maintenance request, I've noticed a lot of times will get changed or declined a day before or a week before, which means you've paid now a really heavy amount and organized and scheduled 50 people to come in for maintenance.
00;11;12;29 - 00;11;34;28
Unknown
And all of that needs to be reorganized. Oh my goodness, that sounds like a oilfield problem if we do that all the time, great. But it's really interesting actually, because oil and sort of utilities and a lot of energy I've noticed, have really similar parallels. Yeah. If you look at it in terms of salary data and industrial, I mean, yeah, right on.
00;11;35;01 - 00;11;52;24
Unknown
And you know, most of the time it's in the most remote unfriendly places to, you know, that's the energy to land. All right. Is the most abundant sources of energy in the world are typically the furthest away from the people that need the energy. And they are there. They don't want it there and told me thing there's yeah, right.
00;11;52;24 - 00;12;14;23
Unknown
They want it close enough but not but directly they're right. Which has been interesting. I moved to Katy nine years ago. And we live in like around an old gas field. And so there's gas pipelines and there's actually a gas well, not far from our house, and stuff. And so it's interesting to see that, but that also gave me a lot of confidence.
00;12;14;23 - 00;12;35;04
Unknown
When I got my generator, I was like, yeah, let's go ahead and install the the hard natural gas line so that we can make make use of this if, the electricity ever goes out again. Yeah. But I think a good starting point with this would be kind of, can you just give a high level for people who aren't familiar with, you know, the wholesale, the back end of the grid?
00;12;35;07 - 00;12;56;08
Unknown
Right. They're just they're a consumer. They don't they pay their monthly thing and they don't ever really comprehend or think about anything from beyond their electric plug. How does you know, let's use Ercot just because that's, where we are. But how does the utility, how does that whole system kind of work? In a nutshell, of course. Great question.
00;12;56;08 - 00;13;15;22
Unknown
And actually it's really interesting because if you look at Ercot and Keiser, I've noticed a quite similar, but if you look at Alabama North, the South, the southern power, for example, doesn't really have a separate ISO. It's all part of the same regulatory body. Right. And so it's very different. Right? Yeah. Yeah. It's very different for them as well to do regulations.
00;13;15;28 - 00;13;45;01
Unknown
Something the way it works right now in a deregulated market like Texas is the regulated market like Texas is. You have all of these utilities which, again, are dealing with a lot of issues already reaching out to a bigger regulatory body like Ercot and having them approve or disapprove a lot of the daily things that they're doing with regards to load management, because Ercot responsibility is being like, okay, hey, all of this is the amount of demand we have.
00;13;45;02 - 00;14;08;01
Unknown
This is what the consumers want. Now, how do we make sure that that is reached? And for them, a lot of times it means managing utilities and saying, hey, you can't do an outage right now. We need you because there it's going to be really hot today and people are going to have their asses on, or there's going to be a storm and we can't risk having a generation source of, accounts job is to keep the lights on to make sure there's no blackouts.
00;14;08;01 - 00;14;40;13
Unknown
And on the utilities, the distribution operators are the ones that are in charge of actually making sure in the field that nothing bad happens. So there's a lot of communication going on constantly between the two of them. As we know, it's like a real time system. I don't think people realize that the prices are updated as frequently as they are and like, it's so dynamic, like it's way more dynamic, I feel like than the oil field is, which is fascinating because like, we complain about all of our problems with it too, like when oil field, you know, emps operators can like capitalize on it with like their natural gas.
00;14;40;13 - 00;15;02;16
Unknown
Right? Because, I mean, they can get outsized returns on their, you know, on their natural gas lines at certain times when they need it. Yeah. Yeah. No, I mean, that was the whole pitch for the Bitcoin miners, right? Yeah. Hey here's a you take a cost, cost center and make it a revenue center. Right. And actually it is getting so much more complex as we talk about it because of a lot of new regulation.
00;15;02;22 - 00;15;25;17
Unknown
This is what we've been spending a lot of our time doing work on is Ferc 881, which essentially is a new regulation that has come and has gone to these ISOs and said, hey, you need to move from standard line ratings or seasonal line readings to hourly line rating. And just to break that down right now, what Keiser does and what Ercot does is they have two line readings a year, which are seasonal.
00;15;25;17 - 00;15;59;18
Unknown
One is for summer, one is for winter that they compare every single outage request or maintenance request that they get. And they say, okay, well, you can do it or you can't do it. Now it's going to be done directly, hourly by 2027 is when it's predicted for some of these ISOs. So it's a really big change, and it's a big change in terms of the utility, because from an operator, dealing with all of the stuff about data centers, I have thousands of people moving in, migrating, having to get them onboarded, making sure that the wires of all of my sort of infrastructure is all.
00;15;59;18 - 00;16;18;22
Unknown
And now I have to deal with an ISO, potentially disapproving or changing my outage request much more often than they would already. Right. And so it's a lot. Yeah, a lot is changing. Wow. Yeah. I didn't even know all that was going on behind the scenes. Well, that's the that's the thing. It's a I mean, it might as well be a trading desk.
00;16;18;22 - 00;16;35;02
Unknown
Like to a certain extent because it's a commodity that's being traded. But you're not just merely around like dollars and cents. I mean, like you've really got like physical infrastructure that will not be done. And like you said, you are you just sitting on it and like then we just had this some of these are very remote operations.
00;16;35;02 - 00;16;57;20
Unknown
It's like, you can't just get this there. Oh, they approved it. Now I have to get there in oh you know like yeah. And it takes everything takes so long. Because even if you look at, getting new transformers, substations, it takes three years to get one procured and built. So not even, but just procured. Yeah. And then you build it out and for building it out, you need to do again an outage to do.
00;16;57;21 - 00;17;20;00
Unknown
Right. That's, that's why all the, the best systems are going after the locations they're targeting are as close to the substation as they can be. So they don't have to build any new infrastructure. Exactly. Which can be a good thing. But when it's, you know, next to an elementary school and next to a creek and in a flood zone, probably not an ideal situation.
00;17;20;02 - 00;17;46;08
Unknown
Not that that is from a direct, thing that I experienced. But, so where do you guys come in? It sounds again, it sounds very similar to even actually worse in some aspects because of the the speed of, of the operations. But, generally speaking, it sounds very similar to, you know, like what we're accustomed to on the upstream side where you've got a bunch of different data coming in at a bunch of different frequencies and a bunch of different stakeholders.
00;17;46;10 - 00;18;12;24
Unknown
There's all this logistics and commute communication that should be happening. It needs to be happening, but I'm sure it's not happening. It's not. And, efficiently or as real time as they might, they would like it to be. Absolutely. Well, there's two main things that we focus on. Number one is reducing as much time as possible, spent on a lot of the admin of it, of getting outage proofs, getting maintenance request requests approved and things like that.
00;18;12;26 - 00;18;32;07
Unknown
We actually have trained a lot of models and a lot of agents on doing these contingency checks from the perspective of an ISO. And so we're able to give you with a really high accuracy, how likely it is that your request would be approved or rejected based on what the load is predicted to be. And we do this on an hourly basis.
00;18;32;07 - 00;18;51;04
Unknown
And so we've been doing research on this for a long, long time. That's one of the things we offer. And that's something we've been set, as we've been told, as a really, interesting approach. Another thing we really look for in the distribution side of things is how we can make things much easier for these operators themselves.
00;18;51;04 - 00;19;21;29
Unknown
So we have a lot of agent services we offer for things like maintenance requests and things like data history on a transformer level and a substation level. And the biggest sort of incentive for us is we want to reduce the interconnection queue, we want to reduce the end consumer price that is being given right now. And the way for us to do it is to spend more time and more money and resources on actually building out more generation and investing in hot tech.
00;19;21;29 - 00;19;48;17
Unknown
Because if you look at the grid right now, it's really unstable to the point where you add a new, based storage, you know, set up or you add a new generation source, it might break. Yeah. And if it does, it's a really hard thing to sort of overcome. And you lose so much money behind it. And so our focus is redirecting resources to all of these really important things by doing a really good job of solving this problem.
00;19;48;24 - 00;20;15;29
Unknown
Yeah. Yeah. So you're not submitting a request, waiting until the week exact to get the rejection and resubmitting it. And yeah, now that makes a ton of sense. There's so much there is so much on the regulatory side that is just so primed for automation. Absolutely. And the rules are defined earlier. That's the thing, right? Like, I mean, any new regulatory workflow that we add in our system.
00;20;15;29 - 00;20;36;18
Unknown
Yeah. I have found that the easiest starting point for it is to let Claude go and scrape that website or that web page and pull the definitions and, you know, the procedure and exactly how you do it. And it's absolutely it has all the context you need. It's, it's yeah. I mean, when Facebook can just like one shot highlights stuff like web okay.
00;20;36;18 - 00;20;54;07
Unknown
They can look at it okay. You know, like something that literally spells out right. Like this internet format. Right. People think agents are this overly complicated thing, but it's just the context required to do a job like it. Exactly is. And if it's a procedure like this where it's like, hey, first you log into this website, then I've always loved, right?
00;20;54;07 - 00;21;17;26
Unknown
Right. I mean, like very deterministic step by step thing. That's what they're built for. That's why I like especially the, you know, the regulatory stuff. There's so many people that their primary job or function is not regulatory, but they have to do regulatory. So and it's like you're paying people way too much money to spend a majority of their time dealing with regulatory stuff that they don't want to do, they don't like to do, but they have to do it right.
00;21;17;28 - 00;21;37;12
Unknown
And instead of spending time doing engineering or whatever, designing the next thing. And so yeah, I think that's a huge there's so much opportunity there. So I know you, you grew up and said, I'm going to do this when I grew up right. This is just growing up. This is my dream. Yeah. Turning data in actionable insights.
00;21;37;15 - 00;22;01;15
Unknown
Absolutely. I'm on that one. You're doing context engineering, actually. So growing up, I grew up in India. Okay. And I came to the US when I was 17 for uni, for my undergrad. And growing up, I always knew energy was something I wanted to work in. We have so many blackouts in India. My co-founder is from Kenya, and so she's very much experienced the same problem.
00;22;01;15 - 00;22;38;07
Unknown
It's what we started off wanting to solve as well is, oh, we want to fix blackouts, which as a sophomore in college is okay, that's a really big yeah, the big goal. That's a big audacious goal or B hag or whatever. Exactly. But I ended up doing chemical engineering because again, growing up, for me, energy was one of those things that you can all of sort of the equity problems and societal problems that I saw growing up had sort of come back to lack of access to reliable energy in terms of having access to the internet, having access to learning more about these things.
00;22;38;07 - 00;22;57;10
Unknown
When you can't have a light on, you can't really think about anything else. Well, that's the whole what Maslow's hierarchy of needs. Like if I'm worried about, like survival or just like being comfortable, I can't tackle like higher levels, right? I'm not worried about regulatory agents. I don't have running water. Exactly. I don't I don't really know about you, but you're not like, you're not the first person I heard this from.
00;22;57;10 - 00;23;17;28
Unknown
There was a, intern we interviewed when I was at university Lands that he was at A&M, but he was doing petroleum engineering, and that's why he chose petroleum engineering. It's like he's like, I grew up in. We didn't have access to reliable energy. And like, this is a wave. Like we, you know, we could provide reliable energy and like, you know, I mean, hospitals didn't have a need to have blackout rolling blackouts at hospitals that you can't provide for people.
00;23;18;01 - 00;23;41;22
Unknown
Right? Absolutely. Well, it's so funny because it's like coming here. It's the complete opposite. We've got it. So at least to the average person, it's so robust that it is so out of sight, out of mind that like now it's the devil, right? Like energy is bad. Yeah. Over time. Right. All this stuff, it's like you realize your entire life runs about five electrons, right?
00;23;41;22 - 00;24;11;08
Unknown
I mean, runs on energy that it's generate. I don't care how it was generate. It was. But it's energy. And I think that's such a big taboo around it now because people don't want to talk about it enough. And all they talk about is data centers. Very important topic to talk about. Not taking away from that, but but I think it's important, very more important in the sense that like this is going to take from or make energy more expensive for the average consumer or in the, in the water side of it, like there are basic needs that people expect and want to be able to have affordably that yeah, this is going to that's
00;24;11;10 - 00;24;28;23
Unknown
it's going to put a strain on it and that way. Yeah. And it's actually really interesting you say that because almost all of the utilities I've talked to, their two biggest priorities are number one, making sure that there's reliable energy access provided. And number two is making sure that it's affordable. Yeah. Like that's such a big priority for these utilities as well.
00;24;28;23 - 00;24;48;25
Unknown
And they're trying hard to achieve it. And so that was an interesting thing for me to learn as well going into it. I mean, yes, as a company, they want it, but they, they also realize like you can't exactly like it's, it's the work that they have to do is really complex. And so our goal again, going into that, going into it with the idea of okay we want to solve blackouts.
00;24;48;25 - 00;25;09;03
Unknown
Yeah. We started off doing a lot of research into this. We've been researching for two and a half, two years. And then we saw a lot of sort of space to grow in this industry. We heard from a lot of people. We were doing research with that we they would like us to develop something for them and actually see, a lot of demand in that, in that field.
00;25;09;06 - 00;25;28;15
Unknown
And so we ended up going full time very recently. It's been 2 to 3 months now, and we signed our first two contracts. So we live in two international utilities, which is very exciting. Thank you. And we just signed our first, U.S design partnership. So we're really looking forward to going in the space now. That's super exciting.
00;25;28;17 - 00;25;58;16
Unknown
Thank you. Thank you. So I just thought about it. You so during Uri, right after I actually wrote a blog and I'm curious on your thoughts on it, or because I'm assuming you've gotten fairly deep on the, like, cloud computing side stuff, too. So it was like basically that designing a grid is similar to how you would design a cloud infrastructure because like, you've got your base, like whether it's nuclear coal, there's like here is an investment fund and like say your EC2 instances would be like natural gas that you can like spin out when you know, on demand when you need it.
00;25;58;18 - 00;26;26;28
Unknown
Right. But then on top of it, like wind and solar are more like your spot instances. Like if they're available, it's great and you can get them cheap and whatever. Exactly. But you can't rely on like they're not always going to be there and they can go away as soon as they can show up. And absolutely. And actually, a lot of the research work we did a while ago, so we were working with this, community in sub-Saharan Africa that was, it was essentially a lot of data on, and sort of in migrant camp and refugee camp.
00;26;27;00 - 00;26;44;06
Unknown
And it was looking at how these mini grids are able to support that. And so much of it is downtime, not because of any other reason, but that's how solar is. And I think it's such an advantage to have on the grid. But it has to be thought about really smartly and built around like you can't afford do that.
00;26;44;09 - 00;27;16;00
Unknown
Like, exactly. And then even just getting back to like say, say oil and gas, like you're not just replacing the energy it creates. There's the whole, you know, downstream petrochemical side of it. You know, the solar will never, you know, people would never realize how much that's it would impact them today. Yeah. I think the big thing with, with the alternative or renewable energy that a lot of the traditional energy people have a problem with is that they look at just the power gen costs of the individual asset and not the entire system.
00;27;16;02 - 00;27;44;00
Unknown
Right. So like solar, great. And it's super cheap. However, to actually leverage that energy, I need a battery storage system as well. That adds tremendous amounts of money to the total cost of the project and all of the other fun stuff. And so it's I think that's where, you know, the, the, the new energy and the traditional energy groups end up butting heads the most because everyone wants to publish their super low number on, on, you know, which one is the best and stuff.
00;27;44;00 - 00;28;14;07
Unknown
And the reality is there is no silver bullet in that, you know, but the the world needs to use the energy that they have available to them wherever they may be, in whatever capacity is available. Right. And so, I think that's the, the bigger thing. But I think, again, when you think of the renewable side, you have to include battery systems now, like that's not an if you want it to be remotely reliable or functional, you need the storage.
00;28;14;09 - 00;28;32;29
Unknown
I would say I think that technology has improved tremendously, like, more so than I thought it would. Yeah. And a handful of years or even like the just like the Tesla battery studies that are coming out. Right. Like once they hit 80%, you know, ten years later, that's basically the bottom of of those batteries. And so it's like, oh, well, that's going to last a hell of a lot longer, right?
00;28;32;29 - 00;28;49;12
Unknown
Yeah. And so, so quickly, who knows what. I'm sure there will be a startup that is recycling Tesla batteries into battery energy storage systems if there isn't already one. Yeah, yeah, there we go. The same thing. Yeah, but but even I will say with the grid it isn't. I don't know if you said Irv I, it's not somewhere like is it in California.
00;28;49;12 - 00;29;07;17
Unknown
Right. Where certain areas have like I said, it's actually cheaper for them to they are like basically giving homes like batteries. Oh yeah. There are places that are trying to. Yeah. Like do a full distributed energy system. There's like I think there's a co-op or two in Texas that's looking at something or trying to incentivize it. I'm not sure what the.
00;29;07;19 - 00;29;25;24
Unknown
Yeah. Because I know some there's been incentive, but I feel like I heard of one. I was I like the, yeah. The, the major like electric company was like giving batteries to like, you know, I mean, it makes selling them for two, you know, in half and simmers just because like, they're like, we now we don't I can take the train off of the grid, right?
00;29;25;26 - 00;29;41;01
Unknown
Right. Like, what does that base power is kind of trying to do that here a little bit. But that's the thing. Like if you gave me, you know, an extra five bucks a month on my utility bill to and then you came and set up the battery like that, why would I not why would I not do that?
00;29;41;01 - 00;30;02;06
Unknown
I would do that all day long so that I at least had a day of backup power for, you know, when inevitably another hurricane comes and shuts down again. Or even just like it built like homebuilders, I think I think that's the problem. People like, the way that it's currently set up is it's, most of the time the battery storage is it's an add on after the house exists.
00;30;02;07 - 00;30;17;11
Unknown
Right? If I had that when the house was built and I could just roll it into my mortgage for $20, oh, $5 a month. Yeah, whatever it would have been, it would've been great because, yeah, now I have I already have my generator. Right. And plug it in. I can use that to, you know, store to the battery.
00;30;17;14 - 00;30;35;09
Unknown
There was that greedy company. Did you ever see pretty. I don't think I said they were, they were trying to leverage the wholesale market in Texas. And so they're the ones that failed during your big time. They. It was great until it wasn't. Right. Well, so their home pitch was if you do it on average, right.
00;30;35;11 - 00;31;02;23
Unknown
You know, you might you're they had a basically a smart contract set up that every 15 minutes would buy wholesale from the grid based off your consumption and demand or expected demand. And so, you know, 80% of the day, you're getting basically free power, right? And then Uri happened and all the, wholesale spot prices exploded. Yeah. All their customers were really pissed for their $500,000 electric bill.
00;31;02;24 - 00;31;32;29
Unknown
Yeah. Right. And so they ended up shutting down, but their one of their like premises was if. Well, we have we have the code and everything already set up so that if you had battery storage at your house, you could set it up so that at a certain time by electricity, you know, from 11 p.m. to 6 a.m. or whatever times, and then drops below a certain price, you to have all this configuration it you buy it, store it to your battery.
00;31;32;29 - 00;31;55;12
Unknown
Then the next day you could just pull off the battery and set that. And I was like, oh, that makes so much right. It's the most like efficient way to utilize. That's what I think the average person also doesn't get is that the energy goes to the grid, whether it's needed or not. A lot of the times. So like at night, it's cheaper for them to just give it away than it is to shut down for two hours and then immediately spin back up.
00;31;55;18 - 00;32;24;29
Unknown
Demand starts back. And so there's there is like there's this optimization. There's an opportunity for efficiency and optimization in the, in that market that I think someone is ultimately going to tease out. And this also sort of brings back to my take on the debate between renewables and, traditional energy sources is also the market will belong to someone who's more innovative right now because it is the time to try out these things and try out giving consumers access to batteries and stuff like that.
00;32;24;29 - 00;32;56;29
Unknown
And I do think the grid is being very innovative and flexible with that right now. I don't think I've ever seen something like this. And so I think it's a great thing to see. But it's also funny you mentioned Uri because I know of a company in California that it signed a PPA with Calpine, and when that happened, the company in California made so much money off of that because it's a they have like they had to and they sort of will be able to sell it back into the grid itself.
00;32;57;02 - 00;33;18;04
Unknown
So it was just like the I don't remember the numbers, but the magnitude of the profit they made on it was insane. Yeah. Yeah. I bet, you know, I've heard from the Bitcoin miners the same thing, right. Like the demand response piece really incentivize them. They'll make more money in a year off of 1 or 2 demand response incidents than they will mining bitcoin the whole year.
00;33;18;07 - 00;33;45;01
Unknown
And so it's like on one hand that's good. You want people to be incentivized to have on demand power that we can revert from, from, you know, intermediate operations to full time critical operations. Right. But you don't want it just like that's that's my problem with the best systems is that they're not there for for the consumer. They're there to make money in the arbitrage.
00;33;45;03 - 00;34;07;24
Unknown
Like that's all they're there for. Right. And so anyway, I don't know how you solve that problem, but let's, let's get into, into the, the tech stack a little bit. So what kind of what's in your stack? Yeah. Of course, right now we use a lot. We use a lot. But the biggest thing is that we.
00;34;07;24 - 00;34;25;09
Unknown
Are you on. Are you operating? Building on. We should be built on cloud. We're on AWS. And we in the utility space, I'm sure, as an oil and gas and a lot of other energy, there's so much compliance that needs to be fulfilled. You have soc2 is the standard one, but then you have ISO certifications you need to do.
00;34;25;09 - 00;34;45;16
Unknown
There's no standards you need to fulfill. And so especially I think building in critical infrastructure is really important that you are in tune with a lot of that. So that's a lot of what we've been working on. We're on AWS right now. For some of our international clients, that's what they prefer, actually, one of them, they want an on site deployment, in Benin.
00;34;45;21 - 00;35;05;11
Unknown
So that's something that we're working on right now. It's exporting. Yeah. So they do that. So like us to that end. Like where did you build it like AWS native like initially. And then do you have to turn it into more containerized? Exactly, exactly. And so we built yeah, we built it out, cloud native. And then we had to make it a bit more siloed.
00;35;05;13 - 00;35;22;24
Unknown
Which is a bit excessive. It's not traditionally what is done in the US. US cloud is great because there's a lot of strict, requirements that you can fulfill based on it, but we're very happy to take the extra step for our clients. Yeah. And maybe easier, knock on wood, maybe easier now than ever with. Oh, yeah.
00;35;22;27 - 00;35;44;20
Unknown
Absolutely. We're actually building out and shipping it so much easier desktop AWS infra emulators now. Okay. I'm like, what is this? This is crazy. But yeah. So compliance is a big thing. A lot of other stuff that we tie in with the very big on our integrations were very diagnostic. Because there's so much going on in the utility space, I don't want to operate.
00;35;44;20 - 00;36;01;18
Unknown
I don't have to deal with another separate tab after having all of these other software that they already have to use that are ATMs that are 25 years old, that you can't really change now because it's been so long, it's going to take a year to switch it. Yeah. And you just don't have the time to. So we tie into a lot of this.
00;36;01;18 - 00;36;24;08
Unknown
Got a data historian data. We tie into OMS systems which is outage management systems we tie into were in the process of getting onboarded with were nervous systems and a lot of the other platforms on that level. And then we also tie into a lot more of the AI models as we were talking about Uri. And I'm sure you guys know about what happened in Nepal recently.
00;36;24;08 - 00;36;49;23
Unknown
Extreme weather events are happening so often, and grid reliability, one of the biggest defenses against it is using AI to predict when and how these climate events are going to be taking place. And so a lot of the work we do as well is evaluating and setting up sort of standards, because it's funny enough. But in this industry, apart from research, there's not actually industry standards for what weather models should look.
00;36;49;23 - 00;37;11;17
Unknown
Energy industry is awful. Any kind of standard, generally speaking. Absolutely. And so that's actually an exciting project we're working on in to see like a whole meteorological side of it as well. Yeah, exactly. I mean perfect sense. Yeah. Yeah. It's not a focus. But I think you have to. Yeah, exactly. And for all of our risk assessment actually we were talking about them being deterministic earlier on.
00;37;11;17 - 00;37;25;03
Unknown
And I can't emphasize more on how much of a design decision that's been, because it's also really hard to convince operators and people in this industry to be like, yeah, don't worry, an AI is going to do it and you can trust it completely. Oh yeah, we talk about it all the time. I mean, we deal with, say, petroleum engineers all the time.
00;37;25;03 - 00;37;40;13
Unknown
They're not going to be okay with, hey, my black box said this and say, well see all the the history. They don't they don't care if it works. Yeah. You could show them that it works, right. It doesn't matter if you can't show them how. Absolutely. And I think that's such a good thing to have in industry as well, because that's the thing.
00;37;40;17 - 00;37;59;12
Unknown
Like that whole industry is like like you need to have that. As much as we talk shit about this, I'm glad it's important. Yeah. And it's really funny. I was talking to an investor or something earlier and they were like, oh, is the accuracy like 95%? And I was like, if it was 95%, we would be like, we wouldn't be alive.
00;37;59;12 - 00;38;25;24
Unknown
We wouldn't be able to survive. Because in a critical infrastructure space that's really, really bad. It needs to be a lot more. Yeah. And so I think we've personally been really lucky because we've been doing a lot of research. We've been able to do a lot of iteration on our product, which I've noticed a lot of the other startups in this field fail just because to get that first pilot, it takes six months of going to compliance and doing all of this, but you're not getting any feedback on your product.
00;38;25;24 - 00;38;47;08
Unknown
And then when you do get the pilot, of course you're going to fail because it's the first time for testing out your product. Yeah. And so on that end, we're quite lucky to have feedback that we're getting consistently from operators. So we're able to build our accuracy up really heavily. Yeah. And so one of our design decisions has been to make decision fatigue sort of disappear.
00;38;47;10 - 00;39;11;05
Unknown
We do a lot of the backend calculations, contingencies very deterministically, with a very, very high accuracy so that an operator doesn't need to spend their time doing a lot of this mapping a lot of these calculations and doing like power solving. Yeah. Models and simulations, they get information that they can trust very accurately, and then they are the ones making the decision at the end of the day.
00;39;11;08 - 00;39;28;22
Unknown
And then we compare it. And we're also building a database on the site. Because another thing that we've noticed really heavily in this industry and in a lot of energy, is the people coming in are not as much as the people leaving. Okay. And so it's also sort of bridging that gap that we're focusing on a bit now.
00;39;28;22 - 00;40;01;09
Unknown
That's a huge piece of this, right, is just how do we capture the domain knowledge in a way that's not sitting down and just interviewing someone until they're dead, right. No, I think that's a huge benefit of of these systems that people don't recognize off the bat, that it's okay. Yes, it's just chat. But what you're chatting about and how you correct it and all your feedback and all of that stuff is very valuable context that can be used to train and build and make things better.
00;40;01;11 - 00;40;22;15
Unknown
So that was one of the first things I was going to bring up. Is that not enough people recognize that these are iterative tools and and things, right. That like listen, everyone testing these things should assume it's going to suck the first probably 2 or 3 iterations. It should get better each iteration. But it's still not going to be perfect.
00;40;22;15 - 00;40;46;29
Unknown
And it won't be because that's how it works. They learn, and so will I want to set the stage because it's a it's a problem that if you're not dealing with it, you don't really understand. But it's like I'm I'm building this Texas equal set up for a client. Right. And it's like, okay, I've got it's got billions of rows in it, multiple tables, some ambiguous relationships across those.
00;40;47;01 - 00;41;17;00
Unknown
And so it's like, okay, I can have an agent sit there and generate, you know, right. Questions based off the structure and stuff. But that's not how you're going to ask for the data from this database. You're not going to say will turn out, you know, filter star table, you're not going to do the sequel. And so at this, like even on the testing side, you first start with like the rawest point, and then you try to get it more semantically closer to a human.
00;41;17;00 - 00;41;38;29
Unknown
Instead of actually using the database names and the table names and stuff. But ultimately you have to have real world examples from the user of how they ask those questions and then what the expect, the expected answer and what the basically the translation to SQL absolute is from that question. And the average person doesn't understand like why when you deploy it, it just doesn't work.
00;41;39;03 - 00;41;57;29
Unknown
The way that they use it. I think it should just always work. It's like, no, you call this. Yeah. There's no I have no idea. It has no right. It doesn't. It knows how to navigate the database all day long. But until you like, until it learns how you call this one field, this acronym or jargon that it's not directly mapped to you.
00;41;58;01 - 00;42;23;07
Unknown
It's not going to understand it, but there. But they learn like my Spotify. Yeah, yeah, yeah, I get them. They get better when I tell it. Good job or bad. Yeah like that. Yes. And it's a lot easier to build them accurately if you have those questions ahead of time. Instead of yeah hoping start from. Yeah. Instead of hoping that the tool, when deployed, automatically knows how to answer all your specific questions 100%.
00;42;23;14 - 00;43;00;16
Unknown
And it's also always different in different companies and different geographies. Oh yeah. Different areas. I mean, across people, across people. Right. And actually, that's one of the biggest roadblocks we ran into when we, first fully began working on this. And the way we went around it, is we did a lot of approval checking. So one of the other features we offer is switching and how we can automate something like switching, which essentially, is if you are doing an outage and you are sending in a crew to do it, you have to make sure that there's a switching line that is already decided on before you go and have a crew,
00;43;00;18 - 00;43;22;06
Unknown
sort of do all the maintenance things or do a switcher, whatever. And writing that itself is long, and the process of checking that is even longer. Okay. And so what we did was we with a lot of our international clients, worked on having more of a comparison thing where we would have our models and agents sort of check what they had done.
00;43;22;06 - 00;43;43;22
Unknown
And eventually we've started having them drafted and then compare it to what an operator would actually write. And now in the next month or so, we're looking to deploy it to the point where it actually writes it. And then we have an operator sort of check it. And so a lot of it we've been noting it's not a thing, is we first want to start with automating a lot of the checking that's being done.
00;43;43;22 - 00;43;58;23
Unknown
Right. Because that again is a lot of work and time on an operator side. And so it's easy enough to give to an AI where you're not completely trusting it, but at the same time, it's improving our model significantly. Yeah, I don't even think about this. But you do an international. I mean, are you having to do this in multiple languages as well?
00;43;58;24 - 00;44;18;22
Unknown
Like the very lucky that not just yet. Okay. That being said, don't sleep on how good the models are at translating. Oh, I have no doubt. But I mean, like, there's, like, it's almost like, you know, well, the say the Bible has been translated, you know, different thing. And there's a telephone like something like this in a certain way, dialect.
00;44;18;22 - 00;44;46;23
Unknown
And like, all of a sudden I'm never capturing the changes, just enough to actually know where, like, a 95% answer is not right. Like, yeah, no, that's very valid. Yeah. We actually have. So we are working with Kenya's national utility Kenya Power right now. And there is some Swahili. So what we also have been looking at right now is maintenance logs that they submit and what data we can get from it because it's not really, arranged in a really nice, pretty scale with sort of all of the yeah, different filters that you can just put on it.
00;44;46;23 - 00;45;04;20
Unknown
It's very messy, which makes sense for any operator in the world. Right now. And so a lot of our work is also translating that and having it converted. But I know we came across this very shortly for one of our problems. And so for that, we were really adamant on getting someone to translate it for us and for us.
00;45;04;20 - 00;45;30;18
Unknown
We're very lucky. My co-founder is Kenyan, so she was able to do it. But yeah, I can't overemphasize how important it is, at least in the beginning, early stages, to have someone go in personally and translate all of that for you. Yeah, because there's so much context that I'm scared of messing with any other translator. When people think like jargon is across an industry, but like jargon can be within a department of a company exactly like their business.
00;45;30;20 - 00;45;56;23
Unknown
Yeah, like many thousand percent. They they are talking about the exact same field in the database, but they are using a different name across three different assets. Right. So it's like those are the areas where you really have to it's important to capture all that stuff 100%. What so what what what models are y'all playing with? What get give some people.
00;45;56;24 - 00;46;17;28
Unknown
Give me the rundown on what you're kind of geeking out over these days. What's exciting? Yeah. Of course, if you look using the open source stuff, is it all cloud code? We are using a lot of open source stuff. A lot of it is again, on the weather model and, especially if you look at, for example, comparing Alabama to Seattle to California and Alabama, storm risk is such a big thing.
00;46;18;02 - 00;46;38;10
Unknown
Seattle and California, you have wildfires coming in. It's such a big thing. And no two open source models are going to be really great at predicting both of them at the same time. And so a lot of what we do is we use, open source models for topology data, which is something that really helps you map out the grid, because a lot of open source models for our, weather and things like that.
00;46;38;10 - 00;46;56;01
Unknown
And a lot of we're actually developing our own one. One of the research papers we have is coming out in a couple of months, which is exciting as well. And a big sort of reasoning for us to do that is, again, context engineering and being able to have our risk assessment sort of cater to every single utility that we're working with.
00;46;56;03 - 00;47;17;05
Unknown
So a lot of it is that a lot of our agent tech work is completely on cloud code, just primarily because we don't want to share it with or we don't want it to rely on an external API that might any time go down. So it's something that we want to be self-reliant in. And so actually what we've also done is we're in the process of network modeling.
00;47;17;05 - 00;47;47;07
Unknown
A lot of grids for different utilities right now. And so it's like 3500 substations. We model it out and we test it before approaching you just to show you sort of the impact that our platform can have on your grid and, and saving you time and money. Right. And so one of I remember we did a recent sort of model for one of the companies in the South, and we noticed that we'd be able to save them almost 18,000 engineer hours just by modeling and simulating what their current work looks like on a model of their grid.
00;47;47;10 - 00;48;04;22
Unknown
Yeah. And so it's yeah, I think it's really, really interesting research to geek out about in this industry. Well, that's the thing. I think people are worried that AI is going to take their jobs, but it's like the first thing it's going to do is make it easier, right? Like instead of you having to go run all those manual calculations, it's just there.
00;48;04;23 - 00;48;25;27
Unknown
It's just that you have more information faster to make a better decision. Exactly. And then you have more time to go in and talk to your consumers. Go in and do more research into the data centers and what that would look like for you. Great to even adopt. Like, there's so much backlog in the industry that AI is the way of catching up is how I like to think of it.
00;48;25;28 - 00;48;49;24
Unknown
Yeah, now it's the scale. That's how you scale for sure. Solutely what what have you been? Have you messed with like, enemies or grok bot or any of, the, the harnesses or anything fun like that? Absolutely. I actually this is crazy. You've mentioned Burmese because I just started using it yesterday when I landed because my friend, from Mexico recommended it heavily.
00;48;49;26 - 00;49;15;12
Unknown
I was using Open Claw a lot before, and for me, it didn't really work out. I realized there's a lot of security loopholes with it, and I'm not very comfortable with having it leak all my information. And so have been messing up meant messing around with things a lot, having been messing around with VMs a whole lot, especially because for us, we have to train a lot of transfer, long running processes, long running processes.
00;49;15;12 - 00;49;43;15
Unknown
And so it's if you look at some of the data sets that we're working with right now, it's tens of thousands of it's basically data every 15 minutes. Yeah. For the last five years. Oh. Oh well it's high frequency doing series data. Right. Exactly. Time series data. And then if you're really heavily training a transformer model and then comparing it to existing models that Nvidia has, and a lot of these other giants hop and then creating our own models based on the transformer model.
00;49;43;15 - 00;49;59;11
Unknown
And it's we actually there's a lot of training work to do. So that also is a very exciting place to look at. And so talk about that because I mean, obviously we've talked a lot about like we're oh, I'm not good at, you know, you actually using data other language models. Right. But like how is this all. How does that work on the back end.
00;49;59;11 - 00;50;29;12
Unknown
Like where you're actually it's helping. Yeah. Are you building like multiple ML models on the time series stuff as well as alarms on the text based side? Or is the LM understand the time series data in some kind of way? Absolutely. So what? I love talking about this, but what we've noticed a lot and I of research has been doing so much work on this as well, is we don't use a on the time series data primarily because we want to run it through a hard model and see what the actual output looks like.
00;50;29;14 - 00;50;45;22
Unknown
One of the biggest things we've been struggling with is bad data, because the data cleaning up, it's great. Big industries have bad data. Yeah, I mean scale especially because like it's kind of immutable. It's like, well, we can't go back and like see what it read. At the time I just said, yeah, but I'm bad value. Yeah.
00;50;45;22 - 00;51;13;15
Unknown
Like, well, right. And it's a really good way of doing, actually my, research just messaged me about an update, with it yesterday, but something that he said that I found really funny was that it's shocking how good a transformer model is at predicting bad data if you run enough of it through the model. And so a lot of what we've noticed is playing around with models to us personally is really exciting to see what kind of configurations work best for the data.
00;51;13;15 - 00;51;37;00
Unknown
We've seen some extremely promising results for even very basic models. And so it's very exciting for us to look at the frontier models in the industry and see how they could help predict and simulate things that haven't been considered before. And a lot of the research we've been doing right now, actually, is with, a company in sub-Saharan Africa that runs a lot of these mini grids.
00;51;37;03 - 00;52;00;25
Unknown
And so for us, we look at it primarily through two sectors. One of them is the harder engineering side, which is where we look at, okay, what does the load generation sort of actually look like. At what time are we doing load shedding. At what time are we going to store these batteries. And a lot of the reliance that is on diesel generators, which again, not the best for the environment.
00;52;00;27 - 00;52;24;20
Unknown
So it's optimizing all of that. And then on the ML it's such an interesting view of how does every single signal correlate to a consumer on a household level. Sure. Which again gives so many inputs that we haven't even been able to consider yet. So a lot of research going on there. Yeah. Because I mean, I know like I think I have actually been in here for the power.
00;52;24;23 - 00;52;44;05
Unknown
Like they were doing stuff on the drilling side where the time series data. But yeah, the ML do the time series, but it recognizes a pattern and does that tied to an action. And then the LM can actually say this pattern right. Goes like now here's the procedures to yeah, you can use action. I mean that's again the LMS OEMs are the great aggregator in my like perspective.
00;52;44;05 - 00;53;01;09
Unknown
Right. You've got all this data. You've got machine data. You've got, you know, this machine messy time series data that's just coming in all the time. And God knows how good or bad it is. And it's definitely going to have random spikes and just noise and shit that on on the basis looks like a signal, but it's just junk.
00;53;01;12 - 00;53;22;23
Unknown
And then at the same time, you also have all the paper documentation that goes along with what is happening on those assets in that time series timeline across that. And so being able to aggregate the data from both of those places and put them in one where it's like, okay, this spike is correlated to this person doing this thing on this job versus this spike is just junk because it's noise.
00;53;22;23 - 00;53;42;10
Unknown
And the sensor freaked out for a second, and we don't need to worry about it. Absolutely right. And so that's that's I still don't think enough people truly recognize the power of what these things can do. When you start, you know, you're running ML models on this time series data. It's looking for all these trends and patterns. It's also running predictive maintenance and all this other stuff.
00;53;42;16 - 00;54;01;28
Unknown
And then meanwhile your data bit you know, your your unstructured data is being, crawled through by agents. That's then correlating back to the timeline that you've built with your time series data. And now all the data is normalized essentially, and contextualized on both sides. Right? It not just that has always been the problem is like if I'm looking at raw data, I don't have the context.
00;54;01;28 - 00;54;24;08
Unknown
And if I'm looking at the context, I don't have the raw data. Yeah. Well, I think you know this. And like even now with LMS, like, again, I'm they're not good at, you know, time series data necessarily. But you know, obviously ML is and but we don't have to be a data scientist anymore. Like, I remember doing like Coursera shit and it's like, oh, we use dplyr and ALM and like all oh, you know, like it was it was cool.
00;54;24;10 - 00;54;46;17
Unknown
Yeah. That was a great. But like but now it's like literally you know the question to ask like you can give it the persona of a data scientist or machine and it knows to go create a test and training split and yeah, you know, run it on random forest XGBoost like, you know, and now compare the results of these, like, you know, 1015, 2020 model whatever it is, and return what the best ones are, which ones are most explainable.
00;54;46;23 - 00;55;07;15
Unknown
Yeah. And yeah. Yeah. That's the most fun part of on the testing. And like R&D side you get done with the run and you're like okay how do we make this better? Yeah, we increase. Guess you haven't even thought about what are what have we done. Yeah. Yeah, absolutely. And I'm so glad you mentioned, the importance of being the importance of being able to decide what is noise and what isn't, because that's such an important thing.
00;55;07;15 - 00;55;29;11
Unknown
And I think we don't emphasize that enough even in research. Because normally the what happens is it's always a look back, right? It's always pro, it's after the fact. Right, right, right. And so like, that makes it so much harder to know any context because God knows how long ago it was. Exactly. And so having it ahead of time makes it easier.
00;55;29;14 - 00;55;48;22
Unknown
Yeah. Earlier you brought up substations and made me think of something fairly totally off, but. And maybe you could get, a contract with the 40 niners. But are you guys aware of, like, the, conspiracy theory with the 40 niners? No, we know that. So it was right around, like, the playoffs last year, I think George Kittle or one of them like towards Achilles or whatever.
00;55;48;24 - 00;56;12;25
Unknown
But this guy had a as a Substack and like had like a, like a three piece like conspiracy theory but saying how like so there's a substation right next to the practice facility. So that's where their locker rooms are. They spend a lot of time there, the training field and like the, you know, according to some research that he was citing or whatever, like based like the waves can like, like reduce the integrity of the soft tissue in your they're basically subs because it's right there.
00;56;12;25 - 00;56;32;18
Unknown
It's emitting like and they have mean no correlation or causation. But they have also had the most soft tissue in the data like five years or so. Oh, the way the data supports that, that could be a problem. Right. But neither the 40 niners or anyone else isn't there. And again, they're not incentivized to do that because. Right.
00;56;32;24 - 00;56;56;03
Unknown
They would of course be right. Like yeah that's horrible. Yeah. But see I was like, maybe you can, you know, create a model for them. Like how many soft tissue injuries they'll have, optimize the level of where you get so much DraftKings or. Oh, yeah, Poly market. Yeah, yeah. Kelsey. Yeah, yeah. I mean, it'd be worth valuing.
00;56;56;06 - 00;57;13;16
Unknown
That's what you see though. I've never heard of that. Yeah, that's good to know. I'd be on the lookout for. No, no. Yeah. Just you can look into that series and then you look at the data and you're like, well, I'm wrong. Yeah, it's definitely correlates. I get into some. Yeah. Kind of a quick little speed around.
00;57;13;16 - 00;57;57;15
Unknown
Yeah. Okay. What is your favorite open source model today? Open source model. Open material. Best for climate in any version. What about, just general or coding? Just general coding. I really like BBC. I've use it a lot. I don't know if that counts as open source. Yeah, yeah, no deep speakers. I think it's also, cloud Code and Codex and all are really great, but they're not I think it's very well I'm playing around with right now is being able to host a model on, commercial machine or a separate server and actually investing in that kind of infrastructure and having it hosted locally in a sort of a closed model.
00;57;57;19 - 00;58;14;20
Unknown
Yeah. Just to sort of protect against when the prices do go high for a lot of these, you know, everyone has an edge against that side of it, right? Yeah. No token cost. And and again, we're at the point of like these, some of these models like if they never got better are extremely good and can do it depend on the work you want.
00;58;14;20 - 00;58;31;03
Unknown
So like if you can lock that in and be like, I know that I'm not going to pay any more for this, like, right now. I mean, that's, a big thing. Oh, I just came out a finally run out there like teams enterprise cloud offering. And so now when you start, Elena, you can you can run just open cloud code running.
00;58;31;03 - 00;58;53;08
Unknown
Oh, on the models. And I told Sam, are one of our IT guys or devs this and, because we're, you know or he's been talking about like okay what are our what our escape hatches. Should there be issues with quad or if, you know, the outages keep getting worse or, you know, they keep cutting back on all the token limits and all of this other stuff.
00;58;53;11 - 00;59;24;28
Unknown
And so I was like, oh, well, they're taking direct aim at that by just having it as the very first thing at the top of the page. Open that thing. Yeah, but it's true. And like deep seek flash is so it's so fast. That is so bad. I think people people underestimate the value of speed with these things because so much of it is like you end up getting to your destination, but through multiple iterations, and the faster you can go through those iterations, the better you can get there and the better it ends up being.
00;59;25;00 - 00;59;58;06
Unknown
And so, you know, while yeah, like fable is, is really good and really smart, it's super expensive and it takes a long time and, you know, you run out of tokens damn near instantly if you're talking about anything remotely big. Yeah. Then what do you like exactly? And so, but I also think people don't like you are talking about, I just did a benchmark yesterday, on some classification workflows using four one mini, five four and five six Luna and four one mini has the least amount of drift.
00;59;58;06 - 01;00;16;19
Unknown
It is the most consistent. And it's the cheapest. Yeah, and it's like, okay, well, I guess what we're going to do that's going to be that for that use case. And yeah, there is a no reason. It doesn't ever have to be better than that. No, it doesn't. And so yeah, and that's the other thing. The newer models, you know, they're pulling more of the controls out of them.
01;00;16;19 - 01;00;38;23
Unknown
So I can't like for Luna, I can't set the temperature to anything because it doesn't have a temperature setting. Yeah. And so when I run it back through the same process doing a classification job, it switches the shit that it classifies for no apparent reason, and there's no way to control it and say no audit history. Right. Like there's, there's yeah, it's so anyway that it's a, it's a fascinating area to be in.
01;00;38;23 - 01;01;01;24
Unknown
And so I yeah, I think so many more people not just from a privacy and cost perspective, but also just from a utilization perspective. Right. Like I was telling one of you, earlier, it's like I will turn the hotspot on in my car and let my laptop keep running while I'm driving home in traffic so that it doesn't disconnect from the job that it's running.
01;01;01;24 - 01;01;17;28
Unknown
Okay, Starbucks. And like, yeah, this is working. Like I put my spot on. I got sick and I'm like, walking to the car with my left. Yeah, yeah. No. Yeah. My laptop sitting open on my console in my truck and but yeah, moving, just having a VM in the cloud that can run those same things allows you to work all the time.
01;01;17;28 - 01;01;31;17
Unknown
It's great. Yeah. No, I've not even started talking to cloud like, well, what can I build? You know, like, build my own. Like. Yeah, I've actually never built a computer. I'll be fun to do anyways. Really? Yeah. Oh yeah, it is fun. But then just have. Yeah, it's a kick ass machine. You saying was running just by a spark?
01;01;31;17 - 01;01;53;25
Unknown
Yeah. Well, I mean, that's the thing now 4K AMD Ryzen stuff whatever. Like. Yeah the new AMD stuff looks interesting I saw speaking of our our stories that we should have started with. I think they bought some company that's running just insane. Like 400 tokens. A second type processing. So like, of the average person, it's about 40 tokens per second.
01;01;53;25 - 01;02;12;26
Unknown
Is what, 40 to 60? Something like that. On a normal foundational model. And this is 400. And like you literally hit enter and it's it just done. Yeah. Yeah. Like holy shit. And so if that is a real thing and they can roll out into their jobs, I think that's, I love how competitive this case is getting across percent ultimately.
01;02;12;28 - 01;02;32;11
Unknown
So is this your first time in Houston or. No, it is my third time in Houston, but my first time alone. Okay. Which is exciting. Yeah. All right. So am I going to to BS question because I've never been to the Bay area. So you're coming here. What's the restaurant that you're going to that you've been wanting to try and Houston and then if I go to the Bay area, what do I need to check out?
01;02;32;14 - 01;02;55;03
Unknown
That's a great question. I actually have no idea. So any recommendations you have for you stay. That would mean any any particular cuisine you're wanting to. I've heard that the seafood has. Good. I don't know if this is amazing until I see it, because the beauty of it is there is not a type of food that you want that you cannot find here.
01;02;55;05 - 01;03;15;19
Unknown
Like, Armenian, which is a very small country that has not have a lot of there's an Armenian church here, there's multiple like Armenia places here that I can get genuine good Armenian food. Right? I mean, I'm trying I mean, there's there's a lot of great seafood places. I'm trying to kick myself in the ass. Well, I mean, there's lots of Tex-Mex seafood.
01;03;15;19 - 01;03;34;23
Unknown
Yeah, but. Yeah. So seafood. I'm thinking, like. I mean, like, it's, it's still there. That little's, oysters, like. It's like it's until, like, I had a really great meal there before, but, like, but, I mean, even, like, if you want some more, like, Cajun style, like seafood, I mean, like, I mean, Landry's and Pappadeaux are hard to beat, like, I mean, there's, there's lots the super nice.
01;03;34;26 - 01;03;53;28
Unknown
Yeah. Maybe six Orleans. But. And then, I mean, some of you probably get plenty of. Yeah, Indian food, but there's a place, guess it's like in Stafford, actually, but it's like supposedly one of the best, Indo-Pak restaurants in North America. I mean, people fly here for it. Know every, well crazy. Yeah, yeah. Okay. I'll definitely check there.
01;03;53;29 - 01;04;12;08
Unknown
Yeah. You should definitely head up. But if you like, if you like barbecue or if you've never had a good barbecue, I mean, what. Yeah. What do you recommend to get together? I'm a Pinkerton's fanboy myself, but, I mean, there's so many truth. I have high standards for barbecue. I will say, after I went, we went to Alabama very recently, and they asked me about, like, from there to, like, Texas.
01;04;12;08 - 01;04;30;22
Unknown
Totally different. Different like, okay, I'm excited to try there. That's all. Yeah. All pork. It's all like sweet generally. Here it's going to be all brisket. It's all beef. That's how you measure the quality of that or beef rib or something. Right. And so it'll be brisket. They'll ask you if you want it dry or fatty.
01;04;30;25 - 01;04;45;13
Unknown
Yeah. All right. What are you seeing closer to downtown or recently? I'm. Yeah. Next to the Galleria. Okay, okay. There's some good stuff out there. Yeah, there's lots of good food over there. There's good food everywhere. Yeah, that's part of the project. But, yeah, if you're coming in, like I'd say, you probably definitely have to get Tex-Mex.
01;04;45;13 - 01;05;07;17
Unknown
And, it's somewhere over here. Tex-Mex is. Yeah. But yeah, it's like, do you want Latin Afro fusion? Because there's a restaurant I can tell you about downtown to go get that, it's like, really? You've got everything. Honestly, the thing I miss the most that I don't feel like we have here now that you're talking about Alabama, is that at least I haven't found it's like a legitimate soul food.
01;05;07;20 - 01;05;38;11
Unknown
Like. Oh, like South. Okay. Like a proper southern, I see, like, blue plate type type thing, but I, I'm sure that that exists. I just have. Yeah, I have it. Yeah. We'll do one more. What what are you most excited about from a technology perspective? Then that can be about, you know, you guys specifically about the, you know, the space in general, but what, what are you most interested in there?
01;05;38;13 - 01;06;06;11
Unknown
I love that question. So much. And I think what I'm most excited about, not just yes, in our space 100%, but also in general, is how physical AI is going to start becoming a lot more decentralized in the sense of how it's going to be more commodified. Because even right now, I don't think physical AI is to the point where you have access to models like ChatGPT or cloud.
01;06;06;11 - 01;06;24;23
Unknown
You can just reach out and you can sort of everyone has a cloud ChatGPT account right now. And I really heavily see at the rate in which physical AI is developing, that in the near future, everyone's going to have access to that kind of technology for themselves, where it's not going to be that hard to build a robot.
01;06;24;26 - 01;06;41;01
Unknown
But then we just, I don't know how I got down this rabbit hole on YouTube with my daughter the other day, but, the new hugging face robot duck popped up and her eyes lit up, and it's like we started watching the demo video and I was like, damn, am I going to get this for you? For, is this what you're doing this year?
01;06;41;04 - 01;07;01;00
Unknown
Because I kind of want one, and I'd love to teach you how to program it, but then it like, I started, you know, an hour later or whatever, I was thinking about it, and I was like, man, this is going to be crazy, right? Like it. I mean, it's 399 and it, you know, it's cute. It sure, it could probably pick up socks or crap off the floor or whatever.
01;07;01;00 - 01;07;27;07
Unknown
It's just got a little mouth. Yeah, but can I throw away your sock rabbit? Yeah. Like yes. Like, hey, that's the first thing we're going to do is we're going to teach it to pick up things off the floor that shouldn't be there. And then it's like, think about actually having access to build your own robots or build your own like computer, or having something that builds it for you or on your home assistant is actually a fine tuned model on all the conversations that you've had.
01;07;27;07 - 01;07;50;03
Unknown
And you can control it. And it's not just Alexa or whatever spying on you all of the time for ads. Yes. Right. Like that was the first. I didn't even realize that. But that was the first, like open source hugging face robot thing that they put out is this cute little Wall-E like robot with little antennas that move around, but it's just a a voice to text, basically a box that sits on your, like, oh, that would be really cool though.
01;07;50;03 - 01;08;11;22
Unknown
Like to I into my Gmail and my calendars and all the family crap and like, like doing laundry, putting up dishes. Yeah. Just so. Yeah, just anyway. So I'm very good at putting laundry in the washer after that. Yeah, right. Not like that. May not me. Oh that's good. The dryer. Dryer. It's not going to fold. At least it'll make it to the dryer eventually.
01;08;11;22 - 01;08;34;28
Unknown
Getting it from the dryer to the dryer is a whole other conversation. Yeah. Things like that. Right. Like, it's going to be wild to see. Exactly. We're not too far from the Jetsons. I know I was telling Colin's that the other day. I was like, when we were talking about the robot, I was like, when we were growing up and stuff, and our parents, back in the old days that we would be having flying cars and.
01;08;35;00 - 01;09;00;04
Unknown
Right. Yeah. Resistance and all of this. Yeah. Like it's start with your life is going to be so much different. Mine is because again, now how we fit on this or something of that size. Yeah. I mean like exactly. And even like taking it further away from home systems is eventually in the future. Kind of like what it was like to get the AI influx of, okay, now I can outsource an entire coding project to my clone or a different model.
01;09;00;07 - 01;09;19;12
Unknown
Eventually in the future, I'll be able to have a drone get all of my own meteorological data for me, instead of having to rely on open radio or any API available. And that's where personally, I'm really excited to see how it'll change, how we do research, how we innovate in the industry, like especially in deep tech as well.
01;09;19;14 - 01;09;44;16
Unknown
I used to be a big biofuel fiend. Before this, I worked at a biofuels club. I joined the biofuel startup. I, yeah, I started my own make your own biofuel. We make our own biofuels, and we connected collect it from the dining halls, and we built up from scratch all chemical engineers. But even things like that, like school project and building out really complex deep tech ventures.
01;09;44;18 - 01;10;06;22
Unknown
How easy is that going to be if you don't need to spend so much time going to a separate EPC firm to get a titanium lead, you know, whatever. Yeah, build out an even the on demand. Like there's all these companies that are spinning up that are doing 3D printed things on demand, 3D printed parts, 3D printed metal.
01;10;06;25 - 01;10;32;10
Unknown
And so it's like, yeah, you need a custom part. You literally can drop the file online and order it and days later. Yeah. It's crazy. Yeah. That's a real thing. Yeah. So yeah, that is very, very exciting to me. Yeah. I'm just worry about the atrophy like the mental atrophy that we all may or may not have like well but like I've thinking about that as well, but like, you know, over time you still have people that work on horses.
01;10;32;12 - 01;10;50;06
Unknown
You just don't have as many. But technology has advanced and yeah, you know, we still have horses, right. Like you're still carriers. Yeah. It's not as my dad used to be one. Yeah. It's not as, you know, common many like my age. Now I'm old. Like, when he turn my age, it's like it's about her, too much.
01;10;50;08 - 01;11;10;01
Unknown
Dude, I couldn't imagine. But if I hunched over a horse for a day, I'd be like, I'd be done. Well, but he he didn't. He's in heat, too. Like, you have to like the t shirt. That's like work. But your dad was jacked. Yeah, that was over. He has all those crazy he was like, yeah, yeah, yeah, yeah.
01;11;10;01 - 01;11;25;16
Unknown
And because, like, he would do, the blacksmithing to like, you know, because like, I mean, some horses like that, I gotta say, it was like a very large, like a Clydesdale type. He would make them. But even like the shoes he would get, he would have to shape in my even so. Yeah, I mean, he's introducing a porch like on his trailer that.
01;11;25;19 - 01;11;49;14
Unknown
Oh there's so much stuff in, in like. Yeah. Oh that's brutal. Yeah. It's hard. No we're just going to all get softer, but we'll keep building shit. Yeah I will say it's got to. It does have a really significant impact because something I've noticed with things like granola, which is the note taker that everyone's using now, people have stopped paying attention in conversations because they're like, oh, my granola record it.
01;11;49;14 - 01;12;08;18
Unknown
Yeah, yeah, I'll go through it later. Exactly like it's fine. It's it's happening. And so they like, turn off their brains in the middle of conversations. And I think well, but was that already happening. And now we just have something to capture it so that we do remember because that happens a lot. Right. Like absolutely. It's I'm very with you on that.
01;12;08;19 - 01;12;38;03
Unknown
We we've been using granola for a long time as well. We because it's not plastic can buy it especially for a dev. If you're doing any kind of custom damage. That's a wonderful workflow. But yeah, it is. It's like I find my I don't know if I'm tuning out more if or I'm just more consciously aware of fact that, hey, this is supporting and therefore I'm now not paying as much attention as I should be, but I think it's also where, like, say, the soft skills and anything authentic is going to have any even more of a value.
01;12;38;03 - 01;12;58;20
Unknown
Absolutely. I completely agree with that. Actually, something that we've been working a lot on is, and something we've noticed a lot is having discourse about this, this podcast even, and coming in person and talking about these topics makes such a big difference. And now we started hosting dinners for operators. We're having our own newsletter come out and highly recommend.
01;12;58;20 - 01;13;20;21
Unknown
That makes such a big difference. Object like in-person presence and discourse over anything. It's crazy that that is like, you know, unique now, right? Exactly, exactly. Easy to meet all the time in person. You're like, that's how it works. Maybe they were talking about this one with the printing press. Like, yeah, I could maybe they're like, you know what?
01;13;20;21 - 01;13;40;05
Unknown
This is gonna be a problem. Like it's all going to be on paper. We have to talk about it. Yeah, right. That's true, actually. That's me. But yeah, I think definitely I like the view of looking at everything that's coming in with an opportunity mindset of, well, whatever. You can't really stop the tech that's coming in. So how can we take it?
01;13;40;10 - 01;13;54;00
Unknown
There is out of the box, like I mean like, yeah, yeah, put it in. But now like I mean, you know, we could do it in America. They're going to keep saying China and Russia could be way behind absolute right. You know, like we have to moving ahead with technology is not even a question. Yeah. But I think that's it's a delicate balance.
01;13;54;00 - 01;14;09;27
Unknown
And I'm excited to see how that goes as well for sure. That's awesome. Amazing. Well, again, thanks for reaching out to us. I mean, like putting putting yourselves on our radar. I mean, you know, sometimes people reach out and you can tell us, like, they want to get kind of salesy with what they're doing. But it's like this was like a genuinely, interesting topic.
01;14;09;27 - 01;14;37;29
Unknown
And just to have someone from out of town come in and, Yeah, yeah. What can we what can we help you with? What are you looking for from from clients or from from, people out there? Do you have any kind of ask me? The biggest ask would be getting in the room, having one conversation. Because what we've noticed is even starting to talk about integrating AI into your system, starting to see what kind of opportunities are available out, then giving us a chance to show what we've built and the kind of history we've had.
01;14;38;01 - 01;15;01;09
Unknown
And the experience we've had built out in this field is more valuable to me than anything else. And so, anyone who's interested in this field and wants to reach out, we'd love to talk. And thank you guys so much. I came across one of your podcasts, I believe, with, someone in the industry, and I was hooked, and I was like, oh, I have to reach out to them because I'm going to use that anyway.
01;15;01;14 - 01;15;18;19
Unknown
Awesome. Such a shame if I don't, but I'm glad you did. That's such a cool, like, full circle thing for me and Bobby. Right? Like, we just do this for fun. We're not trying to make money or become, you know, celebrities or anything off of this week. It's about fun sharing. Yeah, that's that's the whole point. And I think that's why the podcast is of you so authentic and genuine.
01;15;18;19 - 01;15;34;01
Unknown
It's because it is the kind of nerding out about the cool stuff happening in the industry. Yeah, that that's one of the reasons I'm totally cool with not having sponsors, because I don't get to tell us what to do. Exactly. Well that's awesome. And then they can find you on LinkedIn or where else, or yes, you can find me on LinkedIn.
01;15;34;01 - 01;15;58;08
Unknown
I have a Twitter, and those are the best ways to reach me. And what's the company website? It's a video I okay, you spell that real fast. Every RDA I perfect and so res and my email is also at I've already done I so very easy to reach out you know we're fine. Our guys that's. Yes absolutely. Aria thank you so much for for coming on Bobby.
01;15;58;13 - 01;16;13;12
Unknown
Yeah. Good to see me. My question out of here. Thank you so much for having us. Yeah. Thank you. Jealous? You're going to be getting out of town for this? Yeah. Tropical. Whatever like that. But like last week, I started looking as you were hosting a happy hour tomorrow, and I was like, oh, maybe it'll be cooler in Oklahoma City.
01;16;13;12 - 01;16;27;18
Unknown
So I log in to weather.com the way I did it was 108 last week in Oklahoma City, and it was felt like one, don't we here, but it was legitimately 1 to 8 there and like thing tomorrow afternoon because like there's a place social capital I want to do it. Yeah. But then they can rent out the rooftop.
01;16;27;18 - 01;16;55;22
Unknown
It's like it's going to 103 degrees at 5 p.m.. Oh it's great, but yes, I'm better than that warm Oklahoma breeze. Yeah, but it's a dry heat, so. Yeah. Yeah. Right. Awesome guys I appreciate that guys. Yeah. While some may see them as the crazy ones, we see genius. Because the people who are crazy enough to think they can change the world.
01;16;55;24 - 01;16;59;02
Unknown
Are the ones who do good by.