Register for NexusCon 2025 - Now Accepting Abstracts!
Welcome to Nexus, a newsletter and podcast for smart people applying smart building technology—hosted by James Dice. If you’re new to Nexus, you might want to start here.
The Nexus podcast (Apple | Spotify | YouTube | Other apps) is our chance to explore and learn with the brightest in our industry—together. The project is directly funded by listeners like you who have joined the Nexus Pro membership community.
You can join Nexus Pro to get a weekly-ish deep dive, access to the Nexus Vendor Landscape, and invites to exclusive events with a community of smart buildings nerds.
Sign-up (or refer a friend!) to the Nexus Newsletter.
Learn more about the Smart Building Strategist Course and the Nexus Courses Platform.
Check out the Nexus Labs Marketplace.
Learn more about Nexus Partnership Opportunities.
Episode 177 is a conversation with Keith Gipson from Facil.AI and Kenny Seeton from California State University Dominguez Hills.
Episode 177 features Keith Gipson from Facil.AI and Kenny Seeton from California State University Dominguez Hills and is our 15th episode in the Case Study series looking at real-life, large-scale deployments of smart building technologies. These are not marketing fluff stories, these are lessons from leaders that others can put into use in their smart buildings programs. This conversation explores California State University Dominguez Hills’ partnership with Facil.AI to reduce campus energy usage through modern technology. Enjoy!
Monologue from Kenny (0:00)
Introduction (1:25)
Introduction to Kenny (2:49)
Introduction to Keith (5:47)
Tech Stack (6:05)
Implementation (7:25)
Results (17:30)
What’s Next? (31:39)
Lessons Learned (36:17)
Music credits: There Is A Reality by Common Tiger—licensed under an Music Vine Limited Pro Standard License ID: S687981-16073.
Note: transcript was created using an imperfect machine learning tool and lightly edited by a human (so you can get the gist). Please forgive errors!
Kenny Seeton: [00:00:00] The project's important because honestly, we wanna be the first net zero CSU and even better than that, you know, um, we like to lead the, the system. And so we can't do that by running the same way we've always done things. And so helping to push this technology, all the technology really, um, that can reduce our greenhouse gas, make us closer to net zero on our scope one and two, um, is critical.
We can't continue to operate the buildings the way that we did 20 years ago. The power costs too much and. The failure rate, you know, for what if we don't do anything is just too high.
James Dice: Hey friends, if you like the Nexus Podcast, the best way to continue the learning is to join our community. There are three ways to do that. First, you can join the Nexus Pro membership. It's our global community of smart Boeing professionals. We have monthly events, paywall, deep dive content, and a private chat room, and it's just $35 a month.
Second, you can upgrade from the pro membership to our courses [00:01:00]offering. It's headlined by our flagship course, the Smart Building Strategist, and we're building a catalog of courses taught by world leading experts on each topic under the smart buildings umbrella. Third, and finally, our marketplace is how we connect leading vendors with buyers looking for their solutions.
The links are below in the show notes, and now let's go on the podcast.
Brad Bonavida: Welcome to the Nexus Podcast. My name is Brad Bonavita. I am not James Dice. Uh, I am the head of product at Nexus Labs. Um, I'm stepping in for James on this one. Uh, if you're listening, uh, surprise, welcome. You're stuck with me. Buckle up. I'm gonna do my best to keep this thing on the rails, but there's no promises here.
Um, but what isn't changing is the format. Um, so this is the latest podcast in our case study series where we dive into real life, large scale deployments of smart building technologies. We always emphasize real life because we're not here to create some sort of a marketing fluff story. We're here to share real lessons [00:02:00] learned from others who can, you know, tell you about their smart buildings programs, and you can implement that stuff in your smart building programs.
Um, and today we have a story coming out of California State University Dominguez Hills. Uh, C-S-U-D-H is driven by a core commitment to continually reduce campus energy, uh, through modern technology. And one way that they're doing that is with their partnership with FAE AI to improve their chiller performance via fae AI's autonomous optimization solutions.
After a two week ramp up period for faci ai, C-S-U-D-H started consistently seeing improved chiller energy efficiency, uh, also while improving zone temperature comfort in the spaces and equipment served by the chillers. So we're here to tell that story today. Um, so let's start with some context setting.
Um, first, let's have Kenny introduce yourself. Kenny, welcome to the podcast.
Kenny Seeton: Hey, thanks, Brad. My name is Kenny Seton. I am the director of Central Plant Operations and Strategic Energy Projects. Longest title ever. Um, I've been here [00:03:00] since 2011 and my job is everything that has to do with the central plant and energy efficiency.
Brad Bonavida: So since 2011, you've seen quite some stuff happen, I'm sure, at this campus and at this, uh, central plant. Anything you wanna say about what's changed over that span? I
Kenny Seeton: mean, in 2011, we didn't have enough power for the central plant, so we were running direct gas fired absorption chillers to make chilled water burning gas to make cold water.
Uh, we had a, a giant, uh, boiler that with the new energy efficient burner on it, you couldn't run it less than 165 degrees or the condensate would just flood out the front of it. Since that time, we have three electric YMC square york, thousand ton chillers that are magnetic bearing. We have the largest heat pump project on the west coast.
We don't burn gas 95% of the time. And I won't spoil how good our KW efficiency is on the chiller just yet.
Brad Bonavida: All right. We'll get there. Cool. All right, so let's do, uh, this is [00:04:00] kind of a rapid fire set of questions just to set context for the audience. Um, so I know we're talking about the central plant. How does that equate to the amount of buildings that this, uh, affects this project?
Kenny Seeton: Um, so there's about 1.5 million square feet that the central plant, uh, supplies about 13. Main buildings.
Brad Bonavida: Okay. And how, how does that, is there multiple central plants throughout the campus or how does that compare to the rest of the campus size?
Kenny Seeton: No, we have one, one central plant that all the buildings are connected to, with the exception of some trailer buildings.
We're lucky that we have a tunnel system that feeds most of those buildings, so we can walk down through the tunnels and keep an eye on our equipment.
Brad Bonavida: Uh, Kenny, can you talk about the, the vendor team that's been associated with this project internally, externally, who was involved in this chiller plant optimization?
Kenny Seeton: I mean, really it was just Keith and I, um, I, I did involve my team just because to, to do something correctly, you, the whole team needs to be involved, right? They all need to be, have the [00:05:00] same vision so that, you know, they could point things out. But as far as implementation, it was Keith sitting in a chair and, and me telling him what points to talk to.
That was it.
Brad Bonavida: Cool. Okay. And when did you guys start this project?
Kenny Seeton: We probably started it in 2022. Yeah. As far as talking about it, yeah. We actually did the implementation in December of 23.
Keith Gipson: And I want, I wanted to interject, um, as far as this, this partnership, because it really has been a deep collaboration with Kenny.
Like we consider this our r and d. Laboratory, um, customer. So yeah, it was a definitely a, a, a huge collaboration there in real time. So we
Brad Bonavida: built our pro, we built our product here. And well, since, since we got you now, Keith, why don't you go ahead and give yourself a, a brief introduction of who you are.
Sure.
Keith Gipson: Um, Keith Gibson, founder and CEO of facil.ai we're an advanced supervisory control optimization company.
Brad Bonavida: You were talking, how [00:06:00] about how you are at JCI campus? Can you talk Kenny, a little bit about your tech stack and what it looked like before this and what it looks like now?
Kenny Seeton: My tech stack really looks the same.
Right? You know, so we've always had one tech eight hours a month, um, forever, right? Because we, we need that to be able to, you know, to get parts and to do all this stuff. And before Keith, we would use that for, you know, I would have this crazy idea. I'm like, Hey, can you, can you program this? What's it gonna take?
Now I, I have the same thing, but you know, basically I just use 'em for keeping the servers and stuff running and upgrading the software and that kind of stuff. Um,
Keith Gipson: and, and as far as platform wise, but you all, I think, I think another great point, Kenny, is that you had, you, you were one, first of all, Kenny was one of the first in keeping with this tradition.
He was one of the first sky spark. Campuses. So FDD was part of his stack. And then, but we've moved from, um, reactive, you know, 20% trying to find faults and, and chase faults, which nothing wrong with that, that's just not our [00:07:00] business to a 100% autonomous 24 7 optimization profile. So I think that's pretty important.
Brad Bonavida: Got it. So let's talk about that. Let's dive into the deployment a little bit. You guys meet, it sounds like, or you already knew each other, but you start to talk about this, Kenny, you start to get convinced you got nothing to lose here. Uh, there's a lot that Keith could potentially help out with. What's like, step one, what are the phases of this project and the implementation of Fassil on the chillers look like?
Keith Gipson: It's highly automated. I mean, I'd say 100% automated at this point. I mean, of course we built it here, so there was a lot of iterations, but yeah, right now it's literally, uh, one touch operation.
Kenny Seeton: But, but going back to how did it start, right? So I think the first step was I. Uh, getting Keith set up so that he was on the same network as Yes.
As as the Medicis, right? Yeah, that was, that was step one. We set up a desktop in the conference room that is the same as my facilities control specialist desktop. How it works, right? So that then he can see what all the points were multi-home, [00:08:00] Nick and all that stuff. You know, the, the next step for Keith was, was to create that gateway that could just reach in without anything going on.
Pull those data points out. The next step was deciding what the data points were that were important for us. Um, we chose condensing water temperature as as our goal, right? To see, you know, what that would do. It was pull in all those data points, trend all those data points, right? Make sure that everything is reading correctly, and then turn it on and say, Hey look, I want you to look at all these points.
And tell me what it looks like and Keith can go into that part more. So, yeah.
Keith Gipson: Um, that's a, that's a great segue into, I wanna, I wanna, um, describe our kind of transition and growth as far as AI technology. So, version two, version one, we were using another, uh, technology. Um, we are partnered with someone else.
Version two, we came out with our own, this was the end of December, December 26th, 2023. I'll never forget it. Built [00:09:00] this technology and then we started refining. So we first had this concept, it's important to understand the differences of the types of ai. So the first thing that we did was, uh, I would call predictive ai.
So we're trying to, this is like, you know, chat, GPT. Well, that's actually generative ai, but you know, but more like tensor flow and that sort of thing where you're trying to make a prediction as to how you should run the central plant. Um, we quickly outgrew that because a lot of the predictive AI is based on this digital twin concept where you're running all these iterations on a, on a fictitious chiller, I'm gonna call it.
We run our stuff, our AI runs on the real thing. So we're not off here in the cloud doing some simulation on a simulated chiller. We run in real time on the real chiller. So we had to move to the, and I think that's another reason why we're getting the stellar results that we're having. I mean, I always say like, how could you, how could you simulate a central plant?
You might as well try to try to, uh, simulate a tornado. I don't think it's [00:10:00] possible to capture all the nuances. I think that, and
Brad Bonavida: on that point, I think it's one thing I really wanted to do for the listener is like, try to bring to ground the AI that we're talking about here. So we talked about, you guys wanted to focus on condenser water first.
So you're bringing in points from the JCI system that are all related to the condenser loop, right? The temperature, the flow, all those things. What your cooling towers are doing. Try to dumb it down for me. Like what is, you're bringing in all those points and what is the FAE system doing? You know, I understand it's not ru rule based necessarily, but what are all the things that it's doing to test, interpret, make better decisions?
What does that actually look like?
Keith Gipson: Okay. Um, sure. I mean, it's at the business end of heat rejection at the cooling tower. Like the idea you're, what you're trying to do is match the outside the condenser water temperature. Precisely to the outside air wet bulb temperature so that you can have an affinity to reject that heat and transfer it to the atmosphere.
And for the most part, the old rule [00:11:00] of thumb has always been outside air, wet bulb plus five degrees. And so we're, we're, we're doing way better than that because we have a, every five minutes our system calculates the optimal condenser water temperature to, to. You know, assist in that heat rejection process and it, and we're going everywhere from two degrees above, wet above all the way up to like, what, 20 can or something at times.
Oh yeah. I mean, it's just, we just thrown, literally thrown the lid off of what was considered by conventional wisdom. Like everything, one of the old, you know, I'll say Wi Wives tales of, of, of this industry is like, you want to keep your water as cold as possible because that's gonna make the most efficient piece of equipment.
The chiller more efficient. We're not trying to optimize the most efficient piece of equipment in the chain. Don't you wanna optimize the least efficient equipment, which is the cooling tower, which is why you get so much bang for the buck. So you're optimizing put, people are focusing on driving the water [00:12:00] too cold.
That's effectively how we're saving energy because we're not driving the water to cold. Um, it's, it's amazing how much energy it, it takes to heat, to cool down. Uh, that much water. Seven degrees.
Brad Bonavida: Got it. Let me try to say that back to you to make sure that I understood. So you, you're saying that too many people are focusing on trying to optimize the chiller by getting the condenser water loop temperature as cold as possible.
But you're saying that that condenser water loop uses a lot of energy on its own and you guys are also taking into account. Optimizing what energy that cooling tower is using. 'cause it's part of the system as well.
Kenny Seeton: Yes. Yeah. And let me step in there. So, uh, and I think this is an inherent, so Ashra standard 90 19 99, 1 something, 90, yeah.
Ashra 90.1 says that 0.45 kw perton is the best, but it's the chiller. That's all they look at as the chiller. And so I'm telling you that we're averaging [00:13:00]0.45, 0.48 KW per ton for the entire plant. The entire plant is the pumps. It's the cooling tower, it's the condensing water pumps, it's the chiller, right?
And so if you only looked at the chiller, you know, look, I could dump bags of ice in this thing to cool down the condensing water so that I could get cooler water so that my chiller could show a higher efficiency. But how much energy does that take? Right? And so the AI is able to look at all those points, right?
So at Total power. You know, KW per ton, all those things. And at a certain point, the energy that it takes to cool the water at the cooling tower has to merge with the energy that it takes. Intersection. Yeah. That intersection. And so where is that intersection? Right? That's where you wanna run at That perfect sweet spot.
And the AI is able to look at all those points and say, Hey, look, you know what? This is where we're gonna be to, to get that point. Not go over and not go under.
Keith Gipson: By the way, the chiller is always lower than the central plant, so if the plants run at a 0.4, the chiller's at 0.3 or even as low as 0.25, which is just.
Ridiculous efficiency. K [00:14:00] KW per ton is, is what you're talking kw perton, that's
Brad Bonavida: the number. So I, that's where I wanted to double click is, can you guys explain what we're, kind of keep bringing the efficiency back to this KW per ton. How are you guys measuring that? What does that entail to measure it?
Kenny Seeton: So we have, um, when we, when I put the new plant in, when I went from the gas plant to the chiller plant, we put in, uh, all new electrical infrastructure.
We put shark meters in. With cts on every circuit. So I have energy data for every single piece of equipment in our mesis system, we wrote a program that says, Hey, look at this one, this one, this one, add all these up where everything is automated, our flow, our temperature, so you know, tonnages, delta t, the flow, and, and all that.
So all of that stuff is pre-calculated into mesis. It's also because we've been early adopters of skypark. Every one of those data points is brought into skypark. After we had Facil running for a few months, I reached out to my SKYPARK provider. I. And said, Hey, I [00:15:00] need you to do the sanity check for me.
Right before I start posting this stuff on LinkedIn. I need you to tell me, am I looking at this correctly? And they came back and said, wow, yes, this is, this is good. Wait till you get a year's worth of data. And so we've been trending that data for, I could pull up right now, you know, year's worth of data that that just shows all of it and it's all calculated.
We're not touching it anymore. Right? It's all written into the program.
Brad Bonavida: And so is this. It's continually updating the, the actual JCI program or is like what, what is the, um, what are the commands at this point that faci AI is actually writing to the system
Kenny Seeton: right now? So, so we had a program right that said, Hey, wet ball plus five degrees equals condensing water temperature set point.
And now we still have that in case something breaks, um, which hasn't for a long time now, but. Feil is saying, Hey, condensing water temperature, setpoint should be this, [00:16:00] and that's it. And if feil, if, if our heartbeat is good and nobody's hit the kill switch, then that's the command that the. Cooling tower sees.
Keith Gipson: Yeah. Gotcha. So got simple. It just, it just reverts back to what it was doing before outside Air. WaPo plus five. I mean, we can actually tell when Facil, if I'm doing an update or something, Kenny had come up, he says, you know, is facil not running? I'll say, nah, it's down for a couple minutes. Yeah, I could see.
'cause we're, man, we're, we're doing 0.65 kw per ton. What's going on here? Like, we're our own worst enemy now that we're constantly looking at. It's like, oh my God, we're. You know, we're 0.65 when you know, everybody else is thinking they've already arrived when it got there. So.
Brad Bonavida: I wanted to make sure that for our audience, we understood that this is what the term that Nexus Labs uses for all this is advanced supervisory control.
That's what Kenny and Keith are talking about right now. Uh, we have a category in the Nexus marketplace about this. Uh, the key that makes it, you know, considered advanced supervisory control for us is what SCI AI [00:17:00] is doing is actually writing back to the system. So there's no operator here who's changing the set point based on what faci AI says, faci AI is able to change that set point directly and then the system.
You know, VFDs change, pumps change to make that set point a reality. Is that right? Yes. Correct. It's the gen AI, for lack of a word. Exactly. Um, where I was gonna go next is to kind of the operations of this. Kenny, can you talk at all about what, who, who is the team who maintains and operates the chiller and what did that look like before?
And then after facility AI was implemented, did it change at all?
Kenny Seeton: No, nothing changed, right? It's, it's the, the set point is different. Um. And yet it's just, if I were able to do this with programming, right, and do it on out of, in the MESIS system, then I would write a new program and it would say, Hey, this is what the condenser water temperature set point's supposed to be.
Maybe I change it to six and a half degrees above wet bulb or something like that. [00:18:00] Right? And it just works. The only, the only difference is that's gonna be written straight into Mesis. And now AI Feil is saying, do that. I. But IL every five minutes is making a change. And so in Mesis, we wrote into the logic that says, Hey, if you don't see a signal from IL every five minutes, we call it the heartbeat, then run your old program.
Right? And so what's, what's the worst that could happen now? Right? The worst that could happen is I run inefficient because now I know better. Right? Um. But, but we don't lose connectivity. We don't lock up with, you know, a bad set point because, you know, the cloud and computers and all that stuff, um, everything keeps going as it always did.
It doesn't change the building service engineers or the facilities control specialists, what they're doing. Uh, they're still looking for anomalies and for things to go wrong. Just like they always were.
Keith Gipson: Yeah, and that's a great point. I mean, did you see it? I mean, it's, it's so subjective based off [00:19:00] of, of, of Kenny and our, our experience, you know, he said, we just go back to being inefficient.
But that's what the c, that's what the community expects. Like we've already arrived according to Ashray, so not a ding on any of the systems. They're doing a great job. We are making them better.
Brad Bonavida: I like too, that it feels, it feels really clean that it's, uh, like an add-on. Right? You, you were saying it doesn't really change the way anybody's operating.
If it turns off, then things just go back to the way that they were. If it's on, then you're operating more efficiently. So it, it doesn't really affect operations at all. And,
Kenny Seeton: and the truth of it is my team doesn't pay attention to it the same level as I do, so they don't know when it's not working. Yeah, it's, it's me.
I, 'cause I have a big screen monitor that I, you know, that I keep my summary central plan up and I walk by and it's like, look, if, if wet bulb is 53 and the set point is 58, then I know FAE is offline. It's,
Keith Gipson: it's really pathetic at times. We can't even walk by the screen. I've been going through the bathroom or something.
I, I got stop and look at the [00:20:00] screen, you know. But, uh, no, it's, it's completely autonomous in the background and that, and that's way we like it.
Brad Bonavida: Okay. All right. So let's pivot a little bit. Um, I love the story about how you guys started. It's kind of like, you know, the, the perfect team came together to make this thing happen.
My question, and I'd kind of like to hear both of your guys' answers to this, is there was a lot of r and d and figuring it out that it seems like happened here. So if I own a central plant at the next university, or an energy manager, and I'm hearing this. I'm like, okay, great. But like Keith probably isn't gonna come, you know, sit in my central plant for months to like get this figured out.
So like, how does the, how do they apply this? Like now what, what are they gonna do if they don't have a Keith sitting in their office? I have a great
Keith Gipson: answer for that. How did they apply it 30 years ago when I was writing the same sequence in the Commodore 64 hanging on the wall. So I, now I am gonna be a little bit like, let's compare the technology level.
We don't, we, I know how I'm an expert at chillers. I've [00:21:00]commissioned some of the most mission critical central plants in the world. Um, I had a plant at a data center back in the eighties where if they lost any mission, cri any function of the central plant, it cost a million dollars a minute. So I, I've been there, done that.
So if, when I was a controls tech worker for Honeywell or Johnson Controls, nobody asked that question. It was replicated. A central plan is a central plan, is a central plan. Now I, I just have technology that's a thousand times better. So to me that's not, that's like, it's kind of like a false narrative.
And I'll go, I kind of, we did, yes, we did our r and d here, but this thing is repeatable. I. Our stuff takes over. It is an expert on running central plants and it doesn't matter. And we're ramping up
Brad Bonavida: our, our install base right now. So you're saying the next one doesn't need Keith there. You can just take facility AI and bring in the points and let it happen
Keith Gipson: it I would be a service company and that's a good point because there's some folks in the [00:22:00] industry that are nothing more than retrocommissioning and I'll just commissioners and I'll leave it at that.
This is a talk, a software technology AI company. I would, I would have the wor, I would get the mo worst multiplier in, in history if I, I couldn't replicate this in code. Like, that's the whole point. Sure.
Brad Bonavida: And have you, have you been able to replicate it in other chiller plants? Are you, are you moving on to some other ones?
Keith Gipson: Yeah, we've, uh, just landed, um, a few, uh, large central plants in there in implementation right now. So we're doing, uh, USC, several other Cal State. Campuses. I mean, it's just, we got a ux, something coming online in the UAEA district central plant with like 31,000 ton chillers. We're an energy company in the UAE.
So yeah. Um, you know, a month from now, two months from now, we'll have like 30, 40 chillers to prove this out.
Brad Bonavida: Cool. All right. So ke Kenny, same question to you. Like you're, you're the, the central plan operator that, that you don't know, you're not friends with who's, [00:23:00] I don't know, on the other side of the country or something.
They're listening to this, like, how do they get to where you're at?
Keith Gipson: Can I, can I interject real quick? Um, yeah. Uh, close that, put a point, a loop, a bow on that. So. I want to emphasize we were, we are very cautious as a, I'm a serial entrepreneur. This is my seventh startup company in 35 years, I'm very cautious.
Before I bring a a product to market, it's gotta be absolutely bulletproof. So I want to emphasize we didn't start selling our solution until January 1st, 2025. Cool. Okay. Alright. Kenny, go ahead.
Kenny Seeton: What, what I've been telling my, my peers and stuff, right, is like it. It works. Now as an early adopter. I knew what I was signing up for.
Right. You know, monitoring, tweaking, changing. Hey, why don't we do this? Hey, you know what, we need the heartbeat because, you know, the system locked up and locked us at 65 degrees or something. You know, all those changes were [00:24:00] made. It's, it's no different than my, my control software now. Right. You know, they're, they want me to install 14.0.
Right? And it's like, yeah, well you should probably stick to 13.6. Right? Because you don't want to be the guy that goes to zero. But it's the same thing, right? By the time they come out with 15, they're gonna have seven revisions to 14, right? And so it's gonna get better and better. Now, we started off from nothing.
We kind of figured this out. We've been running for a year with very consistent results. Again, I can show you the data. That's as far as it needs to go, right? It's not like I'm telling you, Hey, this works. I can show you that this works. And. I've posted the data on LinkedIn, I've shared it with engineers that, you know, Hey, Ken, can you share us your, you know, the, the CSV file?
Yeah. Here you go. What do you want from us? Well, if you find something that can be even better, you know, let me know. Right. And twice now, we've gotten results back that said, wow, this is really cool. This is what [00:25:00] you're at. You know, and so the, the story is, it's, it's, we're doing it now. Every plant is gonna be different.
You know, we're, we're not naive about that. Right. But the, but the basics of it are right, and so the things that we learned about building the safeguards into the software, right? Mm-hmm. Like, you know, you tell me what your plant is, what you want, the minimum, the maximum. I told Keith that I never wanted to try and get below 60 or try and hit above 90.
We wrote that in now. Yeah. Now the worst case is it locks up at 90. Okay, big deal. What's gonna happen? I'm gonna run 1.5 kw per ton. NI
Brad Bonavida: 90, you're talking about condenser water temperature there?
Kenny Seeton: Yes. Condenser water temperature, sorry. Right. Yeah. You know, and, and so, and depending on what your AIing Right, you know, maybe, you know, the, the next step is, is the air handlers, right?
And it's like, okay, well look, I never wanted to try and hit. 45 degree air, right? Yeah. 55 degree air and I, and we use the same limits built into the AI that we [00:26:00] built into the Medicis. Yeah.
Keith Gipson: Yeah,
Kenny Seeton: that's a good point. And now I can't try and hit a number that's outside of a range that can break things or make people super uncomfortable.
Keith Gipson: So we're quickly, we're quickly moving, um, to, uh, other agents. We had, we had perfected our rooftop unit agent, by the way. So we've been running at a retailer that has 1,850 stores and counting. For three years. Like that's where we were before we jumped over to the chiller.
Brad Bonavida: And, and just to make sure our audience understands what you mean by that, on this project we're talking about a chiller optimization set points there.
You're talking about an HVAC rooftop unit and set points for discharge air temp that you're controlling with FCI
Keith Gipson: discharge and discharge and speed and staging. So yeah, we, we, we've got 5,000 RTUs under our belt and this system saves this particular customer 200,000 to a million dollars a month. So we're getting the same spectacular results.
People said that they're just stupid little rooftop RTUs. You'll never save that much. You'll never move. We're saving [00:27:00] 30, 40%. On rt, on, on so-called stupid RTU. So this AI and I, I'm a, I'm a big car guy, so I'm, and I'm big analogy guy. So this is part like when I had a, I had a Lexus RCF, which is basically a, a luxury $70,000 luxury sports car.
Right. So I was, I was doing some amateur racing and stuff and I was fortunate enough to be, have one of the supercharger manufacturers actually are indeed using my car, right? It was local, Southern California. After they built that supercharger and perfected it on my car and I got it for free because yeah, they could have blew my car up or whatever, but I got it for free.
No one asked the question, if you put this thing on 10,000 more Lexus cfs, if it was gonna work or not. It's a car. They all work the same. So you know, I got a track
Brad Bonavida: record of scaling technology. Got it. Got it. That's a good analogy. I like that. Um, okay, so I know we've, we've talked about the results, but I just wanna make it clear for our audience.
So let's be [00:28:00] succinct. We're talking KW per ton. Can you guys just under make me, uh, give me an explanation of what it was before and what it typically is after? Whether that's an average or seasonally? A
Kenny Seeton: visual average was we were between 0.70 and 1.1. KW per ton is about where we used to run all the time.
Now we run between 0.35 and 0.65. We took a year's worth of data before and a year's worth later, a 15 minute data chopped off the top, chopped off the bottom, created the average, and we ended up that we were wrote it down 0.86 KW per ton average in 2022 and 0.45 KW per average. In 2024. That's
Brad Bonavida: the drop the back moment right there.
And then to, to bring it back to what you guys were saying earlier, that includes not just this chiller, but [00:29:00] the chiller system entirely. So we're talking about the pumps, the VFDs, the, the cooling tower, the whole thing that, that's the cumulative, uh, KW per time. Yeah.
Kenny Seeton: Because that's what it costs to make cold water
Keith Gipson: now.
Kenny Seeton: Right.
Keith Gipson: Speaking of which, there's another important consideration here. So this is not about Rob and Peter to pay Paul, right. This is not about saying I, we had a, a partner of ours, their, their, uh, national partner, HVAC consulting company, mechanical company. They said every time we tried to get below 0.65, Keith, we lost production in tonnage.
Now as you, as you, as you probably know, and maybe the audience doesn't know it, air conditioning, all HVAC and air conditioning equipment is rated in tonnage of output. So you got a 20 ton win. You got a two ton window banger, you got a 20 ton RTU, you got a thousand ton chiller. It's all, and that's just measured output of how [00:30:00] much capacity, how much, how many tons of cooling it can deliver.
So not only are we saving up to 50%, or you know what, 46% or whatever it is, steady state, we're also increasing the output of the chiller by up to 35%. Right? I mean, that's a fair number, Ken. We, we used to, you know, we looked at it yesterday, right?
Kenny Seeton: So yeah, so, so we have a thousand ton chillers. That if I don't bring a second chiller online at 90%, this thing will hit 1300 tons and just stay there all day long.
But better than that is when, um, on, on a different project we were, we were looking at chiller load as a percentage of chiller tonnage, right? And so if you think about it, it's a thousand ton chiller. At 50% load, that should be 500 tons. And that's kind of where it used to be. Now I'll do 500 tons at 0.3, 5.3% load.
[00:31:00] Now that doesn't mean I'm gonna get 1500 tons out of this thing, right? We're still gonna max out at about 1300 tons, but it does mean that my chiller's running so much more efficient that I have no problem doing 900 tons at 75. Percent.
Brad Bonavida: Yeah. And that better, that better output, that better performance.
It's obviously saving you money on your electricity bill, but it's also saying that you're that much less likely to need to add another chiller or just turn on another chiller at any time. Exactly.
Keith Gipson: You just increase your capacity by 30%. Effectively turn a thousand ton chiller into, let's call it a 1300 ton chiller, which is incredible.
Brad Bonavida: Cool. Okay, so what's, uh, what's next, Kenny? What? You got the central plant. You've got the chiller dialed in now, firing a, you know, break in records on the efficiency. So what do you do next?
Kenny Seeton: Uh, the, the next is we're gonna be looking at, um, chiller discharge, uh, water temperature, right? What, what temperature of water do I, what's the optimum temperature of water to be sending out to all those buildings, right?
Yeah, there's an old school people. The [00:32:00] colder the water, the more efficient it is. Well, that goes back to the chiller, right? Yeah. So, because if I run the chill water really cold, then I don't need as much of it. So now my distribution pump gets to slow down. But where's that crossover point at the same time, right?
For, for at the air handler and that kind of stuff. Right? Same exercise. And so we're gonna do the exercise for that. We're gonna look at, um, true optimum start and stop for a chiller, right? If my building runs until 10 o'clock, most people, I. Everything runs until 10 o'clock and then you just flip the switch and everything turns off well, what if FA Seal looks at it and says, Hey, look, you know, all these zones in the building are still at X.
Um, you could shut things off at nine o'clock or nine 30, or maybe you can ramp the air handler down. You don't need that much static pressure. Or maybe you could shut off the chill water completely because you don't need anymore at that time. Right. Um, a lot of little things like that are, are the next step is getting into the air handler, outside air economizer.
Um, letting Theil do it optimally instead [00:33:00] of, I say it's the difference between static settings, right. You know, X plus Y equals Z and that's what you program it for and doesn't matter. Right. But there are a lot of times in the fringes, in the middle, in at the bottom of the range and at the top of the range, that just doesn't work.
Right. We'll let AI do that because that's what we let it do on the condensing water, and it blew our mind. Yeah. And so that's gonna be kind of the next step.
Keith Gipson: Now for, for me at facil speaking at it from a broad view, um, we're moving to many more equipment types and applications. So we, we have two agents right now.
We have the Chill Central Plant. I was almost said Chiller. It is not Central Plant Optimizer, RTU at Scale Optimizer, we're, we've already installed. Um, we've got a, a new customer in Ireland. It's the largest, uh, retail store in Ireland. They've got refrigeration. We're moving to the air side, demand control, ventilation, peak demand mitigation, [00:34:00] boilers.
We're moving. We're gonna have, we're I, I'm projecting that we're gonna have seven more agents, seven or eight more agents by Q3. So we're not, we're not just the one trick we're not gonna be and are not a one trick pony here.
Brad Bonavida: Cool. I think that is the, the AI agents and what they're doing is like by far the most challenging part for any human to grasp here, including myself.
So maybe I'll try one more time to ask it like. Uh, because I don't know if I'm asking the question right, but I'm just, you know, you, you've got this AI agent and it's getting all this data in and it knows that its job is to give the ultimate set point for this one variable. What is the, what is happening there?
Like what is the AI agent doing? Is it testing what works and what doesn't work? Is it learning from history? Keith, can you explain like what is going on in the background to make it make such good decisions?
Keith Gipson: Yeah, this is, this is deep reinforcement learning. Like it's, it's straight DRL and I mean, it's just, we're we, it fires [00:35:00] every, every five minutes, so that's 288 snapshots a day and it's stored all that and I think we're up to.
What, 40, 50,000 rows of data. I mean, we've been running it for close to two years now, so yeah, it never throws a solution away. Remember Ken, we were talking, Kenny's like, well, maybe we should throw the, the bad stuff away. And I was like, well, yeah, that sounds reasonable. And then I was like, wait a minute. I don't think we should throw the bad stuff away.
I think we should keep the bad stuff in there. 'cause it just gets pushed down to the bottom. So yeah, this thing is just, it's a, it's a knob twister at scale. Yeah.
Brad Bonavida: Yeah. So it's constantly looking at what's happened in the past at levels that we can't even quite fathom and seeing how that applies to today's conditions or the develop conditions.
Yeah. And, and a little
Keith Gipson: bit nuance on that, which is the advanced state of our AI technology, I call it, um, uh, variation injection. So we, you don't want to ever, ever think, let the AI think like it's reached, you know, it's, it's, it's done its job and it doesn't get better. It even, it even, we even put in like slight [00:36:00]tweaks to even the settings that work.
I. So that we're constantly pushing, it's constantly experimenting at all times, trying to drive it that much more closer, and we've seen that happen.
Brad Bonavida: Great. Okay, so, uh, kind of starting to wrap here. I'd love to talk about challenges and lessons learned. I mean, it, you know, I think it really, I. Benefits this industry.
When we talk about none of these projects go perfectly, everybody's learning things as they go. Um, maybe we'll go one after another. Kenny, can you think of a challenge or lesson learned that if you could go back in time that you would apply or tell to the next person who's looking to do something like this?
Kenny Seeton: I, I mean, I think for us, because, because it's so new of technology still, um, for me, the, the, the challenge or the lesson learned that, that we implemented was, was the guardrails. And the heartbeat and the kill switch, right? So we wrote into our side of the software, IL has it written into its side. We wrote it into ours also, right?
So [00:37:00] let's, let's be honest, right? People are afraid of this. You know, how do we implement this? We need to make them feel safe. Um. How do you keep AI from running away? Right? You know, you're implementing these things. What happens if it thinks this is the right thing? Or let's say your network connection gets lost or you stop pulling back net points.
There's a, there's some things that can happen, right? And so by creating that heartbeat, that was a huge lesson learned for us, that that saved us, you know? 'cause in the, in the early stage when we were first playing around with it, that happened more than once. Yeah. You know? Um, and so it's like, oh, okay, we need that.
Right. We put the kill switch in because you know, I, I've got a 25 year JCI veteran in there. That's like, you know, he's on board with it. But you know what, if, you know, I'm like, Hey, you know what, just put this in. If you don't like it, just shut it off. If I'm not there, if you can't get ahold of Keith, just shut it off.
Right? Go back to our, our inefficient programming that we had before. You know, I mean, and that's, that's the thing, right? And so, so [00:38:00] those three things. Everybody was comfortable after we got done with that. Right. You know, I was always comfortable with it because like I said, the chillers got enough safeguards to wear, but if you had a different plan with older equipment, maybe that's more important.
Yeah. You know, and so you have to bring your team along on the journey, right? This isn't, this is really cutting edge new stuff, and you can't just throw it on somebody's system that has been happy with, look, I don't have to touch it. And it just runs and nobody complains. You know, and, and so I think those three things are gonna, I think not just FA Seal, but a bunch of AI people hopefully hear this because, you know, they're all trying to shove this down.
A central plant operator's, you know, throat look, our stuff is so great. Let's do this where, what protects meat? Yeah. And, and those were the hurdles truthfully, right? Was how do, how do I make my, my team feel protected?
Brad Bonavida: Right. And so is there, like, you know, on your JCI graphics or whatever your operators are looking at, is there a place [00:39:00] that they can see Face seal AI is on or face SEAL AI is off?
Is there a way that they can tell whether it's on or off
Kenny Seeton: there? There is in a, not in our front end graphics, but in the programming, but in the program. Screen, which anybody can get to if you know where to what, what to click on, or the numbers screen,
Keith Gipson: or if the number is higher than 0.65,
Kenny Seeton: the number is higher than, yeah, there it's,
Brad Bonavida: I think you should get like a cool facile AI neon logo and create a program where when the program, you know, when it's running your faci AI like turns on at the top of the central plan.
When it's not running, it turns off. So.
Kenny Seeton: We could, we could do that. Cool. So Keith,
Brad Bonavida: anything you want to add there to challenges, lessons learned for the next person to do this?
Keith Gipson: Um, yes. Um, so my biggest thing is really, and I, I, this is kind of how I work, um, one of the biggest challenges a lot of times when you're, you're an entrepreneur and you start up a startup company, you're always trying to build the most perfect solution, even as your MVP, right?
We, [00:40:00] we hear people talk about that all the time. Get outside, get out of the vacuum, man. Like stop trying to put this thing down on paper. Find a, you know, I call affectionately the, our Guinea pig, right? He's our Guinea pig. Um, you gotta get out in the real world. Um, and I've been doing that. I'm, I'm, I'm just gonna keep on doing what I've been doing, but trying to do it better because I've learned that this is the best way to bring a, a compelling product that works and is bulletproof, is get there.
Um, I'll give you another story. I, I actually wrote, um, some AI for a, a predictive. I wrote the first predictive policing. Um. Ai, it was actually predicting crime, solving crime. It was featured on CSI, Las Vegas. There's a pretty cool clip of it on our website. But yeah, I didn't know anything about police work, so I got deputized by the Santa Ana Police Department.
I'm, I'm riding in the back seat of the cars. They're kicking in doors, you know, stay here, Keith is the, oh, don't worry, I'm staying here. But I had to immerse [00:41:00] myself in it so I could understand what the cops literally were going through and how we could solve their, their needs. So I'm a real world in the trenches kind of CEO, and that's what I've learned is just keep on doing that and
Brad Bonavida: reinforcing
Keith Gipson: that.
Brad Bonavida: Cool. And then Keith, what are, what are the steps if, if there's someone who owns a central plant who wants to purchase FA Seal and put it in, like what's the first step they should take? What's the second step they should take?
Keith Gipson: Um, so just, yeah, just get ahold of us either, either myself or Danielle. Um, Danielle.
At facil ai.com orKeith@facilai.com we'll, we'll take it from there. And yeah, we prove out all our stuff. You don't pay a penny until you're satisfied that we're bringing value and ROI.
Kenny Seeton: So one of the things I wanna reiterate though, that I, I think a lot of people don't know, is that honestly, yes, until you see proof one month, two months, that it's actually working, you don't pay anything.
Because on newer systems, if [00:42:00] there's no hardware that's involved, Keith just has a, a, a gateway that, you know, he wrote that talks to that system and I can't count how many different alc, couple dozen JCI and couple dozen Honeywell and whatever, right? You know, he goes in there and connects to it and, and sees your back net points and your, and your JCI, medicist, Honeywell, whatever, and it just starts doing its thing, right?
And so. There's not a lot of upfront labor and time and stuff, and so that's where I get back to, you know, what do you got to lose? I. Right. Um, try it.
Brad Bonavida: What do you got to lose? I like that. Cool. I think that's a, a great place to wrap. Um, this is a cool story. We're excited about what you guys are doing, uh, you know, excited to see what continues to come out of this and what C-S-U-D-H does, efficiency-wise moving forward by.
Putting in ai. So thank you, Keith. Thank you, Kenny. It's been a great podcast. Thank you all for our listeners. We kept it on the rails, uh, the whole time. I believe so. Uh, appreciate it. We'll see you at the next one.[00:43:00]
Rosy Khalife: Okay, friends, thank you for listening to this episode. As we continue to grow our global community of change makers, we need your help. For the next couple of months, we're challenging our listeners to share a link to their favorite Nexus episode on LinkedIn with a short post about why you listen. It would really, really help us out.
Make sure to tag us in the post so we can see it. Have a good one.
Welcome to Nexus, a newsletter and podcast for smart people applying smart building technology—hosted by James Dice. If you’re new to Nexus, you might want to start here.
The Nexus podcast (Apple | Spotify | YouTube | Other apps) is our chance to explore and learn with the brightest in our industry—together. The project is directly funded by listeners like you who have joined the Nexus Pro membership community.
You can join Nexus Pro to get a weekly-ish deep dive, access to the Nexus Vendor Landscape, and invites to exclusive events with a community of smart buildings nerds.
Sign-up (or refer a friend!) to the Nexus Newsletter.
Learn more about the Smart Building Strategist Course and the Nexus Courses Platform.
Check out the Nexus Labs Marketplace.
Learn more about Nexus Partnership Opportunities.
Episode 177 is a conversation with Keith Gipson from Facil.AI and Kenny Seeton from California State University Dominguez Hills.
Episode 177 features Keith Gipson from Facil.AI and Kenny Seeton from California State University Dominguez Hills and is our 15th episode in the Case Study series looking at real-life, large-scale deployments of smart building technologies. These are not marketing fluff stories, these are lessons from leaders that others can put into use in their smart buildings programs. This conversation explores California State University Dominguez Hills’ partnership with Facil.AI to reduce campus energy usage through modern technology. Enjoy!
Monologue from Kenny (0:00)
Introduction (1:25)
Introduction to Kenny (2:49)
Introduction to Keith (5:47)
Tech Stack (6:05)
Implementation (7:25)
Results (17:30)
What’s Next? (31:39)
Lessons Learned (36:17)
Music credits: There Is A Reality by Common Tiger—licensed under an Music Vine Limited Pro Standard License ID: S687981-16073.
Note: transcript was created using an imperfect machine learning tool and lightly edited by a human (so you can get the gist). Please forgive errors!
Kenny Seeton: [00:00:00] The project's important because honestly, we wanna be the first net zero CSU and even better than that, you know, um, we like to lead the, the system. And so we can't do that by running the same way we've always done things. And so helping to push this technology, all the technology really, um, that can reduce our greenhouse gas, make us closer to net zero on our scope one and two, um, is critical.
We can't continue to operate the buildings the way that we did 20 years ago. The power costs too much and. The failure rate, you know, for what if we don't do anything is just too high.
James Dice: Hey friends, if you like the Nexus Podcast, the best way to continue the learning is to join our community. There are three ways to do that. First, you can join the Nexus Pro membership. It's our global community of smart Boeing professionals. We have monthly events, paywall, deep dive content, and a private chat room, and it's just $35 a month.
Second, you can upgrade from the pro membership to our courses [00:01:00]offering. It's headlined by our flagship course, the Smart Building Strategist, and we're building a catalog of courses taught by world leading experts on each topic under the smart buildings umbrella. Third, and finally, our marketplace is how we connect leading vendors with buyers looking for their solutions.
The links are below in the show notes, and now let's go on the podcast.
Brad Bonavida: Welcome to the Nexus Podcast. My name is Brad Bonavita. I am not James Dice. Uh, I am the head of product at Nexus Labs. Um, I'm stepping in for James on this one. Uh, if you're listening, uh, surprise, welcome. You're stuck with me. Buckle up. I'm gonna do my best to keep this thing on the rails, but there's no promises here.
Um, but what isn't changing is the format. Um, so this is the latest podcast in our case study series where we dive into real life, large scale deployments of smart building technologies. We always emphasize real life because we're not here to create some sort of a marketing fluff story. We're here to share real lessons [00:02:00] learned from others who can, you know, tell you about their smart buildings programs, and you can implement that stuff in your smart building programs.
Um, and today we have a story coming out of California State University Dominguez Hills. Uh, C-S-U-D-H is driven by a core commitment to continually reduce campus energy, uh, through modern technology. And one way that they're doing that is with their partnership with FAE AI to improve their chiller performance via fae AI's autonomous optimization solutions.
After a two week ramp up period for faci ai, C-S-U-D-H started consistently seeing improved chiller energy efficiency, uh, also while improving zone temperature comfort in the spaces and equipment served by the chillers. So we're here to tell that story today. Um, so let's start with some context setting.
Um, first, let's have Kenny introduce yourself. Kenny, welcome to the podcast.
Kenny Seeton: Hey, thanks, Brad. My name is Kenny Seton. I am the director of Central Plant Operations and Strategic Energy Projects. Longest title ever. Um, I've been here [00:03:00] since 2011 and my job is everything that has to do with the central plant and energy efficiency.
Brad Bonavida: So since 2011, you've seen quite some stuff happen, I'm sure, at this campus and at this, uh, central plant. Anything you wanna say about what's changed over that span? I
Kenny Seeton: mean, in 2011, we didn't have enough power for the central plant, so we were running direct gas fired absorption chillers to make chilled water burning gas to make cold water.
Uh, we had a, a giant, uh, boiler that with the new energy efficient burner on it, you couldn't run it less than 165 degrees or the condensate would just flood out the front of it. Since that time, we have three electric YMC square york, thousand ton chillers that are magnetic bearing. We have the largest heat pump project on the west coast.
We don't burn gas 95% of the time. And I won't spoil how good our KW efficiency is on the chiller just yet.
Brad Bonavida: All right. We'll get there. Cool. All right, so let's do, uh, this is [00:04:00] kind of a rapid fire set of questions just to set context for the audience. Um, so I know we're talking about the central plant. How does that equate to the amount of buildings that this, uh, affects this project?
Kenny Seeton: Um, so there's about 1.5 million square feet that the central plant, uh, supplies about 13. Main buildings.
Brad Bonavida: Okay. And how, how does that, is there multiple central plants throughout the campus or how does that compare to the rest of the campus size?
Kenny Seeton: No, we have one, one central plant that all the buildings are connected to, with the exception of some trailer buildings.
We're lucky that we have a tunnel system that feeds most of those buildings, so we can walk down through the tunnels and keep an eye on our equipment.
Brad Bonavida: Uh, Kenny, can you talk about the, the vendor team that's been associated with this project internally, externally, who was involved in this chiller plant optimization?
Kenny Seeton: I mean, really it was just Keith and I, um, I, I did involve my team just because to, to do something correctly, you, the whole team needs to be involved, right? They all need to be, have the [00:05:00] same vision so that, you know, they could point things out. But as far as implementation, it was Keith sitting in a chair and, and me telling him what points to talk to.
That was it.
Brad Bonavida: Cool. Okay. And when did you guys start this project?
Kenny Seeton: We probably started it in 2022. Yeah. As far as talking about it, yeah. We actually did the implementation in December of 23.
Keith Gipson: And I want, I wanted to interject, um, as far as this, this partnership, because it really has been a deep collaboration with Kenny.
Like we consider this our r and d. Laboratory, um, customer. So yeah, it was a definitely a, a, a huge collaboration there in real time. So we
Brad Bonavida: built our pro, we built our product here. And well, since, since we got you now, Keith, why don't you go ahead and give yourself a, a brief introduction of who you are.
Sure.
Keith Gipson: Um, Keith Gibson, founder and CEO of facil.ai we're an advanced supervisory control optimization company.
Brad Bonavida: You were talking, how [00:06:00] about how you are at JCI campus? Can you talk Kenny, a little bit about your tech stack and what it looked like before this and what it looks like now?
Kenny Seeton: My tech stack really looks the same.
Right? You know, so we've always had one tech eight hours a month, um, forever, right? Because we, we need that to be able to, you know, to get parts and to do all this stuff. And before Keith, we would use that for, you know, I would have this crazy idea. I'm like, Hey, can you, can you program this? What's it gonna take?
Now I, I have the same thing, but you know, basically I just use 'em for keeping the servers and stuff running and upgrading the software and that kind of stuff. Um,
Keith Gipson: and, and as far as platform wise, but you all, I think, I think another great point, Kenny, is that you had, you, you were one, first of all, Kenny was one of the first in keeping with this tradition.
He was one of the first sky spark. Campuses. So FDD was part of his stack. And then, but we've moved from, um, reactive, you know, 20% trying to find faults and, and chase faults, which nothing wrong with that, that's just not our [00:07:00] business to a 100% autonomous 24 7 optimization profile. So I think that's pretty important.
Brad Bonavida: Got it. So let's talk about that. Let's dive into the deployment a little bit. You guys meet, it sounds like, or you already knew each other, but you start to talk about this, Kenny, you start to get convinced you got nothing to lose here. Uh, there's a lot that Keith could potentially help out with. What's like, step one, what are the phases of this project and the implementation of Fassil on the chillers look like?
Keith Gipson: It's highly automated. I mean, I'd say 100% automated at this point. I mean, of course we built it here, so there was a lot of iterations, but yeah, right now it's literally, uh, one touch operation.
Kenny Seeton: But, but going back to how did it start, right? So I think the first step was I. Uh, getting Keith set up so that he was on the same network as Yes.
As as the Medicis, right? Yeah, that was, that was step one. We set up a desktop in the conference room that is the same as my facilities control specialist desktop. How it works, right? So that then he can see what all the points were multi-home, [00:08:00] Nick and all that stuff. You know, the, the next step for Keith was, was to create that gateway that could just reach in without anything going on.
Pull those data points out. The next step was deciding what the data points were that were important for us. Um, we chose condensing water temperature as as our goal, right? To see, you know, what that would do. It was pull in all those data points, trend all those data points, right? Make sure that everything is reading correctly, and then turn it on and say, Hey look, I want you to look at all these points.
And tell me what it looks like and Keith can go into that part more. So, yeah.
Keith Gipson: Um, that's a, that's a great segue into, I wanna, I wanna, um, describe our kind of transition and growth as far as AI technology. So, version two, version one, we were using another, uh, technology. Um, we are partnered with someone else.
Version two, we came out with our own, this was the end of December, December 26th, 2023. I'll never forget it. Built [00:09:00] this technology and then we started refining. So we first had this concept, it's important to understand the differences of the types of ai. So the first thing that we did was, uh, I would call predictive ai.
So we're trying to, this is like, you know, chat, GPT. Well, that's actually generative ai, but you know, but more like tensor flow and that sort of thing where you're trying to make a prediction as to how you should run the central plant. Um, we quickly outgrew that because a lot of the predictive AI is based on this digital twin concept where you're running all these iterations on a, on a fictitious chiller, I'm gonna call it.
We run our stuff, our AI runs on the real thing. So we're not off here in the cloud doing some simulation on a simulated chiller. We run in real time on the real chiller. So we had to move to the, and I think that's another reason why we're getting the stellar results that we're having. I mean, I always say like, how could you, how could you simulate a central plant?
You might as well try to try to, uh, simulate a tornado. I don't think it's [00:10:00] possible to capture all the nuances. I think that, and
Brad Bonavida: on that point, I think it's one thing I really wanted to do for the listener is like, try to bring to ground the AI that we're talking about here. So we talked about, you guys wanted to focus on condenser water first.
So you're bringing in points from the JCI system that are all related to the condenser loop, right? The temperature, the flow, all those things. What your cooling towers are doing. Try to dumb it down for me. Like what is, you're bringing in all those points and what is the FAE system doing? You know, I understand it's not ru rule based necessarily, but what are all the things that it's doing to test, interpret, make better decisions?
What does that actually look like?
Keith Gipson: Okay. Um, sure. I mean, it's at the business end of heat rejection at the cooling tower. Like the idea you're, what you're trying to do is match the outside the condenser water temperature. Precisely to the outside air wet bulb temperature so that you can have an affinity to reject that heat and transfer it to the atmosphere.
And for the most part, the old rule [00:11:00] of thumb has always been outside air, wet bulb plus five degrees. And so we're, we're, we're doing way better than that because we have a, every five minutes our system calculates the optimal condenser water temperature to, to. You know, assist in that heat rejection process and it, and we're going everywhere from two degrees above, wet above all the way up to like, what, 20 can or something at times.
Oh yeah. I mean, it's just, we just thrown, literally thrown the lid off of what was considered by conventional wisdom. Like everything, one of the old, you know, I'll say Wi Wives tales of, of, of this industry is like, you want to keep your water as cold as possible because that's gonna make the most efficient piece of equipment.
The chiller more efficient. We're not trying to optimize the most efficient piece of equipment in the chain. Don't you wanna optimize the least efficient equipment, which is the cooling tower, which is why you get so much bang for the buck. So you're optimizing put, people are focusing on driving the water [00:12:00] too cold.
That's effectively how we're saving energy because we're not driving the water to cold. Um, it's, it's amazing how much energy it, it takes to heat, to cool down. Uh, that much water. Seven degrees.
Brad Bonavida: Got it. Let me try to say that back to you to make sure that I understood. So you, you're saying that too many people are focusing on trying to optimize the chiller by getting the condenser water loop temperature as cold as possible.
But you're saying that that condenser water loop uses a lot of energy on its own and you guys are also taking into account. Optimizing what energy that cooling tower is using. 'cause it's part of the system as well.
Kenny Seeton: Yes. Yeah. And let me step in there. So, uh, and I think this is an inherent, so Ashra standard 90 19 99, 1 something, 90, yeah.
Ashra 90.1 says that 0.45 kw perton is the best, but it's the chiller. That's all they look at as the chiller. And so I'm telling you that we're averaging [00:13:00]0.45, 0.48 KW per ton for the entire plant. The entire plant is the pumps. It's the cooling tower, it's the condensing water pumps, it's the chiller, right?
And so if you only looked at the chiller, you know, look, I could dump bags of ice in this thing to cool down the condensing water so that I could get cooler water so that my chiller could show a higher efficiency. But how much energy does that take? Right? And so the AI is able to look at all those points, right?
So at Total power. You know, KW per ton, all those things. And at a certain point, the energy that it takes to cool the water at the cooling tower has to merge with the energy that it takes. Intersection. Yeah. That intersection. And so where is that intersection? Right? That's where you wanna run at That perfect sweet spot.
And the AI is able to look at all those points and say, Hey, look, you know what? This is where we're gonna be to, to get that point. Not go over and not go under.
Keith Gipson: By the way, the chiller is always lower than the central plant, so if the plants run at a 0.4, the chiller's at 0.3 or even as low as 0.25, which is just.
Ridiculous efficiency. K [00:14:00] KW per ton is, is what you're talking kw perton, that's
Brad Bonavida: the number. So I, that's where I wanted to double click is, can you guys explain what we're, kind of keep bringing the efficiency back to this KW per ton. How are you guys measuring that? What does that entail to measure it?
Kenny Seeton: So we have, um, when we, when I put the new plant in, when I went from the gas plant to the chiller plant, we put in, uh, all new electrical infrastructure.
We put shark meters in. With cts on every circuit. So I have energy data for every single piece of equipment in our mesis system, we wrote a program that says, Hey, look at this one, this one, this one, add all these up where everything is automated, our flow, our temperature, so you know, tonnages, delta t, the flow, and, and all that.
So all of that stuff is pre-calculated into mesis. It's also because we've been early adopters of skypark. Every one of those data points is brought into skypark. After we had Facil running for a few months, I reached out to my SKYPARK provider. I. And said, Hey, I [00:15:00] need you to do the sanity check for me.
Right before I start posting this stuff on LinkedIn. I need you to tell me, am I looking at this correctly? And they came back and said, wow, yes, this is, this is good. Wait till you get a year's worth of data. And so we've been trending that data for, I could pull up right now, you know, year's worth of data that that just shows all of it and it's all calculated.
We're not touching it anymore. Right? It's all written into the program.
Brad Bonavida: And so is this. It's continually updating the, the actual JCI program or is like what, what is the, um, what are the commands at this point that faci AI is actually writing to the system
Kenny Seeton: right now? So, so we had a program right that said, Hey, wet ball plus five degrees equals condensing water temperature set point.
And now we still have that in case something breaks, um, which hasn't for a long time now, but. Feil is saying, Hey, condensing water temperature, setpoint should be this, [00:16:00] and that's it. And if feil, if, if our heartbeat is good and nobody's hit the kill switch, then that's the command that the. Cooling tower sees.
Keith Gipson: Yeah. Gotcha. So got simple. It just, it just reverts back to what it was doing before outside Air. WaPo plus five. I mean, we can actually tell when Facil, if I'm doing an update or something, Kenny had come up, he says, you know, is facil not running? I'll say, nah, it's down for a couple minutes. Yeah, I could see.
'cause we're, man, we're, we're doing 0.65 kw per ton. What's going on here? Like, we're our own worst enemy now that we're constantly looking at. It's like, oh my God, we're. You know, we're 0.65 when you know, everybody else is thinking they've already arrived when it got there. So.
Brad Bonavida: I wanted to make sure that for our audience, we understood that this is what the term that Nexus Labs uses for all this is advanced supervisory control.
That's what Kenny and Keith are talking about right now. Uh, we have a category in the Nexus marketplace about this. Uh, the key that makes it, you know, considered advanced supervisory control for us is what SCI AI [00:17:00] is doing is actually writing back to the system. So there's no operator here who's changing the set point based on what faci AI says, faci AI is able to change that set point directly and then the system.
You know, VFDs change, pumps change to make that set point a reality. Is that right? Yes. Correct. It's the gen AI, for lack of a word. Exactly. Um, where I was gonna go next is to kind of the operations of this. Kenny, can you talk at all about what, who, who is the team who maintains and operates the chiller and what did that look like before?
And then after facility AI was implemented, did it change at all?
Kenny Seeton: No, nothing changed, right? It's, it's the, the set point is different. Um. And yet it's just, if I were able to do this with programming, right, and do it on out of, in the MESIS system, then I would write a new program and it would say, Hey, this is what the condenser water temperature set point's supposed to be.
Maybe I change it to six and a half degrees above wet bulb or something like that. [00:18:00] Right? And it just works. The only, the only difference is that's gonna be written straight into Mesis. And now AI Feil is saying, do that. I. But IL every five minutes is making a change. And so in Mesis, we wrote into the logic that says, Hey, if you don't see a signal from IL every five minutes, we call it the heartbeat, then run your old program.
Right? And so what's, what's the worst that could happen now? Right? The worst that could happen is I run inefficient because now I know better. Right? Um. But, but we don't lose connectivity. We don't lock up with, you know, a bad set point because, you know, the cloud and computers and all that stuff, um, everything keeps going as it always did.
It doesn't change the building service engineers or the facilities control specialists, what they're doing. Uh, they're still looking for anomalies and for things to go wrong. Just like they always were.
Keith Gipson: Yeah, and that's a great point. I mean, did you see it? I mean, it's, it's so subjective based off [00:19:00] of, of, of Kenny and our, our experience, you know, he said, we just go back to being inefficient.
But that's what the c, that's what the community expects. Like we've already arrived according to Ashray, so not a ding on any of the systems. They're doing a great job. We are making them better.
Brad Bonavida: I like too, that it feels, it feels really clean that it's, uh, like an add-on. Right? You, you were saying it doesn't really change the way anybody's operating.
If it turns off, then things just go back to the way that they were. If it's on, then you're operating more efficiently. So it, it doesn't really affect operations at all. And,
Kenny Seeton: and the truth of it is my team doesn't pay attention to it the same level as I do, so they don't know when it's not working. Yeah, it's, it's me.
I, 'cause I have a big screen monitor that I, you know, that I keep my summary central plan up and I walk by and it's like, look, if, if wet bulb is 53 and the set point is 58, then I know FAE is offline. It's,
Keith Gipson: it's really pathetic at times. We can't even walk by the screen. I've been going through the bathroom or something.
I, I got stop and look at the [00:20:00] screen, you know. But, uh, no, it's, it's completely autonomous in the background and that, and that's way we like it.
Brad Bonavida: Okay. All right. So let's pivot a little bit. Um, I love the story about how you guys started. It's kind of like, you know, the, the perfect team came together to make this thing happen.
My question, and I'd kind of like to hear both of your guys' answers to this, is there was a lot of r and d and figuring it out that it seems like happened here. So if I own a central plant at the next university, or an energy manager, and I'm hearing this. I'm like, okay, great. But like Keith probably isn't gonna come, you know, sit in my central plant for months to like get this figured out.
So like, how does the, how do they apply this? Like now what, what are they gonna do if they don't have a Keith sitting in their office? I have a great
Keith Gipson: answer for that. How did they apply it 30 years ago when I was writing the same sequence in the Commodore 64 hanging on the wall. So I, now I am gonna be a little bit like, let's compare the technology level.
We don't, we, I know how I'm an expert at chillers. I've [00:21:00]commissioned some of the most mission critical central plants in the world. Um, I had a plant at a data center back in the eighties where if they lost any mission, cri any function of the central plant, it cost a million dollars a minute. So I, I've been there, done that.
So if, when I was a controls tech worker for Honeywell or Johnson Controls, nobody asked that question. It was replicated. A central plan is a central plan, is a central plan. Now I, I just have technology that's a thousand times better. So to me that's not, that's like, it's kind of like a false narrative.
And I'll go, I kind of, we did, yes, we did our r and d here, but this thing is repeatable. I. Our stuff takes over. It is an expert on running central plants and it doesn't matter. And we're ramping up
Brad Bonavida: our, our install base right now. So you're saying the next one doesn't need Keith there. You can just take facility AI and bring in the points and let it happen
Keith Gipson: it I would be a service company and that's a good point because there's some folks in the [00:22:00] industry that are nothing more than retrocommissioning and I'll just commissioners and I'll leave it at that.
This is a talk, a software technology AI company. I would, I would have the wor, I would get the mo worst multiplier in, in history if I, I couldn't replicate this in code. Like, that's the whole point. Sure.
Brad Bonavida: And have you, have you been able to replicate it in other chiller plants? Are you, are you moving on to some other ones?
Keith Gipson: Yeah, we've, uh, just landed, um, a few, uh, large central plants in there in implementation right now. So we're doing, uh, USC, several other Cal State. Campuses. I mean, it's just, we got a ux, something coming online in the UAEA district central plant with like 31,000 ton chillers. We're an energy company in the UAE.
So yeah. Um, you know, a month from now, two months from now, we'll have like 30, 40 chillers to prove this out.
Brad Bonavida: Cool. All right. So ke Kenny, same question to you. Like you're, you're the, the central plan operator that, that you don't know, you're not friends with who's, [00:23:00] I don't know, on the other side of the country or something.
They're listening to this, like, how do they get to where you're at?
Keith Gipson: Can I, can I interject real quick? Um, yeah. Uh, close that, put a point, a loop, a bow on that. So. I want to emphasize we were, we are very cautious as a, I'm a serial entrepreneur. This is my seventh startup company in 35 years, I'm very cautious.
Before I bring a a product to market, it's gotta be absolutely bulletproof. So I want to emphasize we didn't start selling our solution until January 1st, 2025. Cool. Okay. Alright. Kenny, go ahead.
Kenny Seeton: What, what I've been telling my, my peers and stuff, right, is like it. It works. Now as an early adopter. I knew what I was signing up for.
Right. You know, monitoring, tweaking, changing. Hey, why don't we do this? Hey, you know what, we need the heartbeat because, you know, the system locked up and locked us at 65 degrees or something. You know, all those changes were [00:24:00] made. It's, it's no different than my, my control software now. Right. You know, they're, they want me to install 14.0.
Right? And it's like, yeah, well you should probably stick to 13.6. Right? Because you don't want to be the guy that goes to zero. But it's the same thing, right? By the time they come out with 15, they're gonna have seven revisions to 14, right? And so it's gonna get better and better. Now, we started off from nothing.
We kind of figured this out. We've been running for a year with very consistent results. Again, I can show you the data. That's as far as it needs to go, right? It's not like I'm telling you, Hey, this works. I can show you that this works. And. I've posted the data on LinkedIn, I've shared it with engineers that, you know, Hey, Ken, can you share us your, you know, the, the CSV file?
Yeah. Here you go. What do you want from us? Well, if you find something that can be even better, you know, let me know. Right. And twice now, we've gotten results back that said, wow, this is really cool. This is what [00:25:00] you're at. You know, and so the, the story is, it's, it's, we're doing it now. Every plant is gonna be different.
You know, we're, we're not naive about that. Right. But the, but the basics of it are right, and so the things that we learned about building the safeguards into the software, right? Mm-hmm. Like, you know, you tell me what your plant is, what you want, the minimum, the maximum. I told Keith that I never wanted to try and get below 60 or try and hit above 90.
We wrote that in now. Yeah. Now the worst case is it locks up at 90. Okay, big deal. What's gonna happen? I'm gonna run 1.5 kw per ton. NI
Brad Bonavida: 90, you're talking about condenser water temperature there?
Kenny Seeton: Yes. Condenser water temperature, sorry. Right. Yeah. You know, and, and so, and depending on what your AIing Right, you know, maybe, you know, the, the next step is, is the air handlers, right?
And it's like, okay, well look, I never wanted to try and hit. 45 degree air, right? Yeah. 55 degree air and I, and we use the same limits built into the AI that we [00:26:00] built into the Medicis. Yeah.
Keith Gipson: Yeah,
Kenny Seeton: that's a good point. And now I can't try and hit a number that's outside of a range that can break things or make people super uncomfortable.
Keith Gipson: So we're quickly, we're quickly moving, um, to, uh, other agents. We had, we had perfected our rooftop unit agent, by the way. So we've been running at a retailer that has 1,850 stores and counting. For three years. Like that's where we were before we jumped over to the chiller.
Brad Bonavida: And, and just to make sure our audience understands what you mean by that, on this project we're talking about a chiller optimization set points there.
You're talking about an HVAC rooftop unit and set points for discharge air temp that you're controlling with FCI
Keith Gipson: discharge and discharge and speed and staging. So yeah, we, we, we've got 5,000 RTUs under our belt and this system saves this particular customer 200,000 to a million dollars a month. So we're getting the same spectacular results.
People said that they're just stupid little rooftop RTUs. You'll never save that much. You'll never move. We're saving [00:27:00] 30, 40%. On rt, on, on so-called stupid RTU. So this AI and I, I'm a, I'm a big car guy, so I'm, and I'm big analogy guy. So this is part like when I had a, I had a Lexus RCF, which is basically a, a luxury $70,000 luxury sports car.
Right. So I was, I was doing some amateur racing and stuff and I was fortunate enough to be, have one of the supercharger manufacturers actually are indeed using my car, right? It was local, Southern California. After they built that supercharger and perfected it on my car and I got it for free because yeah, they could have blew my car up or whatever, but I got it for free.
No one asked the question, if you put this thing on 10,000 more Lexus cfs, if it was gonna work or not. It's a car. They all work the same. So you know, I got a track
Brad Bonavida: record of scaling technology. Got it. Got it. That's a good analogy. I like that. Um, okay, so I know we've, we've talked about the results, but I just wanna make it clear for our audience.
So let's be [00:28:00] succinct. We're talking KW per ton. Can you guys just under make me, uh, give me an explanation of what it was before and what it typically is after? Whether that's an average or seasonally? A
Kenny Seeton: visual average was we were between 0.70 and 1.1. KW per ton is about where we used to run all the time.
Now we run between 0.35 and 0.65. We took a year's worth of data before and a year's worth later, a 15 minute data chopped off the top, chopped off the bottom, created the average, and we ended up that we were wrote it down 0.86 KW per ton average in 2022 and 0.45 KW per average. In 2024. That's
Brad Bonavida: the drop the back moment right there.
And then to, to bring it back to what you guys were saying earlier, that includes not just this chiller, but [00:29:00] the chiller system entirely. So we're talking about the pumps, the VFDs, the, the cooling tower, the whole thing that, that's the cumulative, uh, KW per time. Yeah.
Kenny Seeton: Because that's what it costs to make cold water
Keith Gipson: now.
Kenny Seeton: Right.
Keith Gipson: Speaking of which, there's another important consideration here. So this is not about Rob and Peter to pay Paul, right. This is not about saying I, we had a, a partner of ours, their, their, uh, national partner, HVAC consulting company, mechanical company. They said every time we tried to get below 0.65, Keith, we lost production in tonnage.
Now as you, as you, as you probably know, and maybe the audience doesn't know it, air conditioning, all HVAC and air conditioning equipment is rated in tonnage of output. So you got a 20 ton win. You got a two ton window banger, you got a 20 ton RTU, you got a thousand ton chiller. It's all, and that's just measured output of how [00:30:00] much capacity, how much, how many tons of cooling it can deliver.
So not only are we saving up to 50%, or you know what, 46% or whatever it is, steady state, we're also increasing the output of the chiller by up to 35%. Right? I mean, that's a fair number, Ken. We, we used to, you know, we looked at it yesterday, right?
Kenny Seeton: So yeah, so, so we have a thousand ton chillers. That if I don't bring a second chiller online at 90%, this thing will hit 1300 tons and just stay there all day long.
But better than that is when, um, on, on a different project we were, we were looking at chiller load as a percentage of chiller tonnage, right? And so if you think about it, it's a thousand ton chiller. At 50% load, that should be 500 tons. And that's kind of where it used to be. Now I'll do 500 tons at 0.3, 5.3% load.
[00:31:00] Now that doesn't mean I'm gonna get 1500 tons out of this thing, right? We're still gonna max out at about 1300 tons, but it does mean that my chiller's running so much more efficient that I have no problem doing 900 tons at 75. Percent.
Brad Bonavida: Yeah. And that better, that better output, that better performance.
It's obviously saving you money on your electricity bill, but it's also saying that you're that much less likely to need to add another chiller or just turn on another chiller at any time. Exactly.
Keith Gipson: You just increase your capacity by 30%. Effectively turn a thousand ton chiller into, let's call it a 1300 ton chiller, which is incredible.
Brad Bonavida: Cool. Okay, so what's, uh, what's next, Kenny? What? You got the central plant. You've got the chiller dialed in now, firing a, you know, break in records on the efficiency. So what do you do next?
Kenny Seeton: Uh, the, the next is we're gonna be looking at, um, chiller discharge, uh, water temperature, right? What, what temperature of water do I, what's the optimum temperature of water to be sending out to all those buildings, right?
Yeah, there's an old school people. The [00:32:00] colder the water, the more efficient it is. Well, that goes back to the chiller, right? Yeah. So, because if I run the chill water really cold, then I don't need as much of it. So now my distribution pump gets to slow down. But where's that crossover point at the same time, right?
For, for at the air handler and that kind of stuff. Right? Same exercise. And so we're gonna do the exercise for that. We're gonna look at, um, true optimum start and stop for a chiller, right? If my building runs until 10 o'clock, most people, I. Everything runs until 10 o'clock and then you just flip the switch and everything turns off well, what if FA Seal looks at it and says, Hey, look, you know, all these zones in the building are still at X.
Um, you could shut things off at nine o'clock or nine 30, or maybe you can ramp the air handler down. You don't need that much static pressure. Or maybe you could shut off the chill water completely because you don't need anymore at that time. Right. Um, a lot of little things like that are, are the next step is getting into the air handler, outside air economizer.
Um, letting Theil do it optimally instead [00:33:00] of, I say it's the difference between static settings, right. You know, X plus Y equals Z and that's what you program it for and doesn't matter. Right. But there are a lot of times in the fringes, in the middle, in at the bottom of the range and at the top of the range, that just doesn't work.
Right. We'll let AI do that because that's what we let it do on the condensing water, and it blew our mind. Yeah. And so that's gonna be kind of the next step.
Keith Gipson: Now for, for me at facil speaking at it from a broad view, um, we're moving to many more equipment types and applications. So we, we have two agents right now.
We have the Chill Central Plant. I was almost said Chiller. It is not Central Plant Optimizer, RTU at Scale Optimizer, we're, we've already installed. Um, we've got a, a new customer in Ireland. It's the largest, uh, retail store in Ireland. They've got refrigeration. We're moving to the air side, demand control, ventilation, peak demand mitigation, [00:34:00] boilers.
We're moving. We're gonna have, we're I, I'm projecting that we're gonna have seven more agents, seven or eight more agents by Q3. So we're not, we're not just the one trick we're not gonna be and are not a one trick pony here.
Brad Bonavida: Cool. I think that is the, the AI agents and what they're doing is like by far the most challenging part for any human to grasp here, including myself.
So maybe I'll try one more time to ask it like. Uh, because I don't know if I'm asking the question right, but I'm just, you know, you, you've got this AI agent and it's getting all this data in and it knows that its job is to give the ultimate set point for this one variable. What is the, what is happening there?
Like what is the AI agent doing? Is it testing what works and what doesn't work? Is it learning from history? Keith, can you explain like what is going on in the background to make it make such good decisions?
Keith Gipson: Yeah, this is, this is deep reinforcement learning. Like it's, it's straight DRL and I mean, it's just, we're we, it fires [00:35:00] every, every five minutes, so that's 288 snapshots a day and it's stored all that and I think we're up to.
What, 40, 50,000 rows of data. I mean, we've been running it for close to two years now, so yeah, it never throws a solution away. Remember Ken, we were talking, Kenny's like, well, maybe we should throw the, the bad stuff away. And I was like, well, yeah, that sounds reasonable. And then I was like, wait a minute. I don't think we should throw the bad stuff away.
I think we should keep the bad stuff in there. 'cause it just gets pushed down to the bottom. So yeah, this thing is just, it's a, it's a knob twister at scale. Yeah.
Brad Bonavida: Yeah. So it's constantly looking at what's happened in the past at levels that we can't even quite fathom and seeing how that applies to today's conditions or the develop conditions.
Yeah. And, and a little
Keith Gipson: bit nuance on that, which is the advanced state of our AI technology, I call it, um, uh, variation injection. So we, you don't want to ever, ever think, let the AI think like it's reached, you know, it's, it's, it's done its job and it doesn't get better. It even, it even, we even put in like slight [00:36:00]tweaks to even the settings that work.
I. So that we're constantly pushing, it's constantly experimenting at all times, trying to drive it that much more closer, and we've seen that happen.
Brad Bonavida: Great. Okay, so, uh, kind of starting to wrap here. I'd love to talk about challenges and lessons learned. I mean, it, you know, I think it really, I. Benefits this industry.
When we talk about none of these projects go perfectly, everybody's learning things as they go. Um, maybe we'll go one after another. Kenny, can you think of a challenge or lesson learned that if you could go back in time that you would apply or tell to the next person who's looking to do something like this?
Kenny Seeton: I, I mean, I think for us, because, because it's so new of technology still, um, for me, the, the, the challenge or the lesson learned that, that we implemented was, was the guardrails. And the heartbeat and the kill switch, right? So we wrote into our side of the software, IL has it written into its side. We wrote it into ours also, right?
So [00:37:00] let's, let's be honest, right? People are afraid of this. You know, how do we implement this? We need to make them feel safe. Um. How do you keep AI from running away? Right? You know, you're implementing these things. What happens if it thinks this is the right thing? Or let's say your network connection gets lost or you stop pulling back net points.
There's a, there's some things that can happen, right? And so by creating that heartbeat, that was a huge lesson learned for us, that that saved us, you know? 'cause in the, in the early stage when we were first playing around with it, that happened more than once. Yeah. You know? Um, and so it's like, oh, okay, we need that.
Right. We put the kill switch in because you know, I, I've got a 25 year JCI veteran in there. That's like, you know, he's on board with it. But you know what, if, you know, I'm like, Hey, you know what, just put this in. If you don't like it, just shut it off. If I'm not there, if you can't get ahold of Keith, just shut it off.
Right? Go back to our, our inefficient programming that we had before. You know, I mean, and that's, that's the thing, right? And so, so [00:38:00] those three things. Everybody was comfortable after we got done with that. Right. You know, I was always comfortable with it because like I said, the chillers got enough safeguards to wear, but if you had a different plan with older equipment, maybe that's more important.
Yeah. You know, and so you have to bring your team along on the journey, right? This isn't, this is really cutting edge new stuff, and you can't just throw it on somebody's system that has been happy with, look, I don't have to touch it. And it just runs and nobody complains. You know, and, and so I think those three things are gonna, I think not just FA Seal, but a bunch of AI people hopefully hear this because, you know, they're all trying to shove this down.
A central plant operator's, you know, throat look, our stuff is so great. Let's do this where, what protects meat? Yeah. And, and those were the hurdles truthfully, right? Was how do, how do I make my, my team feel protected?
Brad Bonavida: Right. And so is there, like, you know, on your JCI graphics or whatever your operators are looking at, is there a place [00:39:00] that they can see Face seal AI is on or face SEAL AI is off?
Is there a way that they can tell whether it's on or off
Kenny Seeton: there? There is in a, not in our front end graphics, but in the programming, but in the program. Screen, which anybody can get to if you know where to what, what to click on, or the numbers screen,
Keith Gipson: or if the number is higher than 0.65,
Kenny Seeton: the number is higher than, yeah, there it's,
Brad Bonavida: I think you should get like a cool facile AI neon logo and create a program where when the program, you know, when it's running your faci AI like turns on at the top of the central plan.
When it's not running, it turns off. So.
Kenny Seeton: We could, we could do that. Cool. So Keith,
Brad Bonavida: anything you want to add there to challenges, lessons learned for the next person to do this?
Keith Gipson: Um, yes. Um, so my biggest thing is really, and I, I, this is kind of how I work, um, one of the biggest challenges a lot of times when you're, you're an entrepreneur and you start up a startup company, you're always trying to build the most perfect solution, even as your MVP, right?
We, [00:40:00] we hear people talk about that all the time. Get outside, get out of the vacuum, man. Like stop trying to put this thing down on paper. Find a, you know, I call affectionately the, our Guinea pig, right? He's our Guinea pig. Um, you gotta get out in the real world. Um, and I've been doing that. I'm, I'm, I'm just gonna keep on doing what I've been doing, but trying to do it better because I've learned that this is the best way to bring a, a compelling product that works and is bulletproof, is get there.
Um, I'll give you another story. I, I actually wrote, um, some AI for a, a predictive. I wrote the first predictive policing. Um. Ai, it was actually predicting crime, solving crime. It was featured on CSI, Las Vegas. There's a pretty cool clip of it on our website. But yeah, I didn't know anything about police work, so I got deputized by the Santa Ana Police Department.
I'm, I'm riding in the back seat of the cars. They're kicking in doors, you know, stay here, Keith is the, oh, don't worry, I'm staying here. But I had to immerse [00:41:00] myself in it so I could understand what the cops literally were going through and how we could solve their, their needs. So I'm a real world in the trenches kind of CEO, and that's what I've learned is just keep on doing that and
Brad Bonavida: reinforcing
Keith Gipson: that.
Brad Bonavida: Cool. And then Keith, what are, what are the steps if, if there's someone who owns a central plant who wants to purchase FA Seal and put it in, like what's the first step they should take? What's the second step they should take?
Keith Gipson: Um, so just, yeah, just get ahold of us either, either myself or Danielle. Um, Danielle.
At facil ai.com orKeith@facilai.com we'll, we'll take it from there. And yeah, we prove out all our stuff. You don't pay a penny until you're satisfied that we're bringing value and ROI.
Kenny Seeton: So one of the things I wanna reiterate though, that I, I think a lot of people don't know, is that honestly, yes, until you see proof one month, two months, that it's actually working, you don't pay anything.
Because on newer systems, if [00:42:00] there's no hardware that's involved, Keith just has a, a, a gateway that, you know, he wrote that talks to that system and I can't count how many different alc, couple dozen JCI and couple dozen Honeywell and whatever, right? You know, he goes in there and connects to it and, and sees your back net points and your, and your JCI, medicist, Honeywell, whatever, and it just starts doing its thing, right?
And so. There's not a lot of upfront labor and time and stuff, and so that's where I get back to, you know, what do you got to lose? I. Right. Um, try it.
Brad Bonavida: What do you got to lose? I like that. Cool. I think that's a, a great place to wrap. Um, this is a cool story. We're excited about what you guys are doing, uh, you know, excited to see what continues to come out of this and what C-S-U-D-H does, efficiency-wise moving forward by.
Putting in ai. So thank you, Keith. Thank you, Kenny. It's been a great podcast. Thank you all for our listeners. We kept it on the rails, uh, the whole time. I believe so. Uh, appreciate it. We'll see you at the next one.[00:43:00]
Rosy Khalife: Okay, friends, thank you for listening to this episode. As we continue to grow our global community of change makers, we need your help. For the next couple of months, we're challenging our listeners to share a link to their favorite Nexus episode on LinkedIn with a short post about why you listen. It would really, really help us out.
Make sure to tag us in the post so we can see it. Have a good one.
Head over to Nexus Connect and see what’s new in the community. Don’t forget to check out the latest member-only events.
Go to Nexus ConnectJoin Nexus Pro and get full access including invite-only member gatherings, access to the community chatroom Nexus Connect, networking opportunities, and deep dive essays.
Sign Up
Nexus Labs exists to move the smart buildings industry forward, without the fluff.
On Wednesday, we send out the renowned (and free) Nexus Newsletter.