Change Makers: Alex Klaessig, Halcyon Chief Data Officer

View Full Transcript

Hill Vaden:

All right, welcome back to EnergyCents, an S&P Global Commodity Insights podcast covering all topics on the intersection of energy and finance. This is your host Hill Vaden, here with your other host Sam Humphrey.

Sam, how's it going?

Sam Humphrey:

It's all right. Thanks, Hill. How are you?

Hill Vaden:

I am doing all right. You're just back from travel and you've got a cold or something?

Sam Humphrey:

Yeah, it's basically the plague. I sound really croaky and I probably coughed through most of that episode, so sorry people listening. If you're on your treadmill, I'm just going to be coughing in your ear.

Hill Vaden:

Well, people on the treadmill should feel grateful because they're working hard and not coughing.

Sam Humphrey:

Yeah, exactly.

Hill Vaden:

Right, because they're healthy. Well, we have just spoken with, as a part of our Change Makers series, we've just spoken with Alex Klaessig, who many of our listeners will remember from episode 39 of EnergyCents, where he was on as a former colleague of ours. Today he's talking with us as the chief data officer at Halcyon. Halcyon is a new company, I guess a startup, it's a couple years old. But they're developing AI technology really focused on energy, and today on North America power and gas markets. Combing utilities, and Public Utility Commission, and other information, so it's a very specific AI application, which seems to have a lot of great interest and a lot of great needs.

Can you flag some things for the listeners to put their ears out for?

Sam Humphrey:

Yeah, sure. I think this is a particularly interesting one because obviously we're talking about AI a lot at the moment. Most people will use ChatGPT or the equivalent in their day-to-day. This technology is really focused at helping researchers get the most out of very niche information, and distilling it down to something usable, which is not something that is available on a lot of those platforms at the moment. This has a lot of applications within the energy sector. I think he did mention that it's now across all 50 states in the US, which is key because then it allows for cross-state comparison. I think part of what they were doing is levelizing some of the language that is used between states to try and enable that comparison, to make it a lot easier for the folks researching.

It's very interesting. I think there's a lot of applications that this could be used for going forwards. Like you said, it's a couple of years old and they're just starting to build this out now. It's just a really fascinating approach to research I think that could be really beneficial.

But what about you, what did you think?

Hill Vaden:

Yeah, I'll echo that. I think it's pretty exciting to hear, one, to hear someone like Alex involved in this because he's coming to this from the perspective of having been an energy analyst so he knows what the output that would be valuable would be. Two, the idea of sector-specific AI, I think it'd make a lot of sense. You can find correlations in anything. I remember years ago, I think somebody found correlations between oil price and Astros' win streaks. If one is limiting and working with AI in a very domain-specific way, there's probably opportunity for more robust information. It's an interesting thing to watch and a good conversation, so we'll hand off now.

Sam Humphrey:

Enjoy.

Hill Vaden:

All right, Alex, thank you so very much for joining us today of an EnergyCents Change Makers discussion to talk about your work with Halcyon. I guess first, I'd like to understand the name Halcyon, which I associate as the salad day, that memory of good times or when things were great, which is an interesting approach to a company focused on AI and the future. How did you come up with that name?

Alex Klaessig:

My founder boss wanted to have a name like Kingfisher because the application that we've developed is really good at finding information or finding things that are buried deep underwater, as it might be for the kingfisher example. We talked him out of that. We said, "Hey, that's just something that you remember from being a kid when you lived in South Africa." He was like, "Oh, the halcyon days," and it just stuck. It seemed like a great name to give some optimism for what is sometimes seen as a dehumanizing technology I guess. Hopefully, it's working.

Hill Vaden:

Halcyon, there's a double entendre, right? It is a kingfisher, right?

Alex Klaessig:

Yeah.

Hill Vaden:

You've got the kingfisher and you've got the salad days. Great, okay. What does this kingfisher salad days Halcyon technology, can you describe the technology and describe the company for us?

Alex Klaessig:

Yeah, yeah, absolutely. Back when I worked at S&P, or it was IHS back in the day, I was on the power team and I spent a fair amount of time looking through really long regulatory documents. I remember we had this one colleague who spent maybe a month digging through integrated resource plans for electric utilities to pull out pertinent information. When my friend Bruce who founded the company asked me, "Hey, what do you think we can do with these LLM technologies?" I said that would be the type of document that I would want someone else to read. If we can make the software do that for us, that's amazing. We've developed a suite of tools that harvest all of this regulatory information that's out there and make it very easily for analysts at a company like S&P to get through it very fast.

Sam Humphrey:

Who are you building this for? Are you building this for researchers, are you building this for companies? Are you building this for government officials looking at navigating the energy transition? What's your target audience?

Alex Klaessig:

I guess, like all good founders, I'm building it for myself. Our client profiles are the energy consulting world, developers, investors. There is this enormous ecosystem of people that are working with electric and natural gas utilities and they really represent, I don't know, maybe four or five percent of the US economy. There is a lot of information that's going on there and we want to be part of that ecosystem. From energy insight firms like yourselves to investment banks who want to know what's going on in rate cases for the investor-owned utilities that they might have a portfolio on.

Hill Vaden:

Okay. You're focused, am I right, on North America, or US documents? I guess also, can you back us up a little bit to founding, 2023, '24, and some of the investor and where you are from a business perspective?

Alex Klaessig:

Yeah, sure. I've known Bruce for maybe 15, 20 years. My wife and his wife were really good friends way back in the day. He had a lot of experience at Google and at Twitter. When Twitter got turned upside down, I guess is a way to say it, he was one of the people who got turned upside down. I was with him maybe a year or two later and I was asking him what he was going to do. He said to me, "Well, I want to bring this LLM and search technology that I'm an expert at into the energy world, and I'm hoping that you can help me." I said, "Well, that's great. What's the idea? What can I help you with?" He said, "Voluntary carbon markets." I was like, "Oh, that's a terrible idea." He's like, "Well, what would you do?" That's when I told him the story of my colleague with the integrated resource plans.

Just to back you up a little bit on the source material or some of the hypothesis of the company. These LLM and foundational models, they require a data corpus and at least, in our technology, it's a research-augmented generation. Which means that we have a body of knowledge that we access to find information, and then we bring you back an answer, and we use a large language model like OpenAI or ChatGPT to put it in a native language wrapper. Ideally, we can port it across many different languages, but right now we are focused in North America.

The natural gas and electric companies that I mentioned earlier are geographic monopolies. They are granted a fiefdom, and inside of that fiefdom there is a regulated utility or a public utility commission that really governs the way they operate. They go to the PUC every time they want more money or need to increase rates to make sure that it is justifiable and prudent. The PUC gives them a chance to earn a regulated rate of return. The amount of paperwork that these proceedings take, it's overwhelming. You have a lot of consultants and other folks that are very specialized in following these things at a very minute level.

The advances in AI technology have really made it so that there has been, I don't know, a democratization of this information. We can get through it fast, we can go across different states. The underlying mission that Halcyon is under is to make this information more accessible so that we can help out the trillions of dollars that are being deployed in this industry. Because I don't know about you, I live in Massachusetts, we already have high enough electricity prices.

Sam Humphrey:

If you're a customer, I'm just trying to get this right in my mind, the platform you provide is, I guess like an OpenAI chatbot that you can ask questions to and it will go and find the answers, is that correct? To some extent?

Alex Klaessig:

Part of that functionality, you ask a native language query, but it's also a bit more about understanding the data that's out there. There is a lot of contextualization that the platform provides.

One of the things that we are differentiating ourselves with is specialization. In the oil and gas industry or in the electric business, specialization allows you to go deep into an area and ideally be much better at it than a more general resource. We do all sorts of things to make it so that not only is the data catalog that we have focused in the electric and natural gas utilities, but there's a lot of stuff that we can do with the queries that we send over to our internal systems to provide additional context.

If I'm thinking about individual utilities, I might want to know when their electric generating units are retiring. If I say, "Hey, tell me when Avista's units are retiring?" On the back end we can say we have a list of units that Avista owns, let's send those along to the LLM so that the LLM has great context for providing a good answer. Identifying that kind of friction point is really where we're at as a company right now. We are building software to make it so that people have better and better search results.

That's one of the real nice parts about my co-founder Bruce's background, with the Google and the Twitter expertise, he's very good at building systems that aggregate information and aggregate customers' preferences to understand what's relevant. That's really what it is, applying search in the energy realm to understand what's relevant and give you back a great answer.

Hill Vaden:

Can you talk a little bit more about the user experience? Would this be like a Twitter or a Google in the sense that everybody on the planet has access to it? Or given its sector specialization or domain specialization, yes, it's going to appeal to a certain persona, but how are you guys targeting that persona differently than say, Twitter with a broadcast, or what it's called now, X?

Alex Klaessig:

The ecosystem that we live in with information is that it's really becoming much more of a commodity. We have a bunch of things in terms of feeds and alerts that allow people to understand quite rapidly what's going on with various topics around the 50 different public utility commissions across the nation. That's something that you basically have to give away for free because other people are doing that. In order to be sticky, you have to draw people in with a superior offering of how you distribute that.

What we're able to do with the AI is not only say, "Hey, this new document has come out from the California Public Utility Commission," but we can also summarize it for you. We can give you a quick little paragraph with the who, what, when, where, and why. That allows you to determine quite rapidly, "Is this something I even want to read?" That type of user preference or user discovery is what we're doing on the non-paywall side.

Then on the paywall side, there's the query application. Then when you're bringing a new technology to market, especially in the software realm, there's friction for the customer. I think we've all had this experience with ChatGPT where you go in and you type in an answer, and it gives you a nice result, but it's 80% of what you want. You think to yourself, "Well, I could do better," so you just move on. We're trying to develop the answer that's a 95, a 99 percent. We want that A-level answer. We have a help desk that allows people to say, "Hey, I want to know this body of knowledge. What is FERC compliance with their Rule 2023?" We can go into the back end and not only run batch queries against all of the relevant dockets, but we can also QA that so that people don't get awkward responses or returns that don't make sense.

Sam Humphrey:

I guess with a lot of these interactive AI models that require you to ask a question, is there the functionality to upload documents to this to then have analysis or that context? Say if there's a new regulatory announcement, for example. I'm trying to word this better. Can you upload that and then see how that would then compare to, say the rest of the states across the US? Or have I missed the point there?

Alex Klaessig:

No, no, no. That's absolutely one of the-

Sam Humphrey:

Okay, fine.

Alex Klaessig:

[inaudible 00:15:16]. There, this is an area, again, where you're right at the frontier of what the technology can do. You run into, I don't know, the frontier a lot. A single line of thought is often times 50 or 60 different queries. If a new piece of legislation or regulation comes out and you want to compare it to the old one, you need a rubric of questions that you've asked the old one, a rubric of questions that you've asked the new one, and then you need to do a comparison of the two. That is the type of end-to-end functionality that the technology does not do yet. You need to build additional technology in order to be able to have a completely automated system like that. That's where we're at right now, is building that technology.

Here, you run into the examples that you see from other AI companies. There's companies like Hebbia and Harvey. Harvey does legal and Hebbia does financial infos. Essentially, what these companies did is they found one work stream that they would automate from end-to-end and do exactly what you're talking about. Be able to have a single question that you put in with a new document or some new documentation and get a legal document out of it. That is where we are going with our technology right now.

Hill Vaden:

Where are you in the company formation process? I think you've got venture backing, or are customers able to interact with you through an app today, through a website? Can you talk about the ecosystem?

Alex Klaessig:

Yeah, absolutely. We have venture backing. We got our seed round of funding back in October 2023. We basically had to build out our infrastructure from there. There are 50 different PUCs and maybe 10 different RTOs, and FERC, and the SEC. We just finished our last state crawler today, so we can ring the bell on Vermont being the last one done. This is really important, just in terms of a go-to-market strategy because when we talk to investment banks and big energy insight firms, the response that we got was that unless we had nationwide coverage, we're really not a great data provider. Just because there's this mental effort for the customer to say, "Well, if it's in Ohio, I know they don't have that state." That was one of the big pieces of feedback we got for our go-to-market strategy.

Then there's also just what is the business model that you're going to go to? Are you going to do seats? A lot of people like a seat model. I personally like an enterprise model, just because I'm terrible with accounting. Just navigating that interplay between what are the clients' needs and what are the things that are going to make us a good company is something that we've been working toward.

In terms of the way that we've been going to market with the clients, initially we were doing these design partner programs. It was very much we want exposure to the clients' workflow so that we can understand where we need to make improvements to even have a chance. We're well beyond that point now. Now, we have all 50 states. Now, we're at the point where we're going through and just understanding how much information can we extract and what is the best way that we can do that.

You asked about the query box and the query tool earlier, Samantha, and it doesn't work quite like ChatGPT. But for an analyst, they don't want to ask 50, a million questions about all of these different documents. They want [inaudible 00:19:04] done for them, or they want it already done so they can just download it and have it. That's where we're at now. We have this incredibly catalog with millions upon millions of documents. They have a wealth of information. Now we're going through and identifying, "Hey, data centers in Virginia, how big is this facility? What load is it bringing online?" And that type of information, once it's organized, it's very valuable. Right now, we're organizing.

Sam Humphrey:

You mentioned the other two companies back there, was it Heber and Harvey? I like that people used Hs, being what, finance and legal?

Alex Klaessig:

Yeah.

Sam Humphrey:

Can you explain some of the unique differences for the energy transition sector compared to those offerings? They're not just a lift and copy and paste model, I take it.

Alex Klaessig:

Oh, no. Each one of these state PUCs has their own jargon and their own terminology, and just collecting the stuff is a real massive undertaking. The websites are antiquated, and sometimes they fight us off, and sometimes they crumble our warm embrace. All of the different types of file names and standardizing it, it might be comments in one state or it might be intervener suggestions in another. Just making those crosswalks is a massive undertaking. A big part of it has just been creating normalized views that allows you to make these cross-state comparisons.

Before AI and the advent of LLMs, that was very difficult to do at scale. They have done it for the various different legal proceedings that are in the various states. We're doing a very similar thing for the energy sector, but just the concepts and the way the information flows is different.

Hill Vaden:

I'm going to ask what feels like a Simpson's did it question that, if you're familiar with the South Park episode from many years ago, where the South Park was having trouble coming up with ideas that the Simpsons hadn't already done. Why would I want a sector-specific AI tool if Google's already done something, or ChatGPT's got these very large non-sector-specific AI bots or LLM? Pitch me the value of something that is more narrowly intelligent, so to speak.

Alex Klaessig:

Okay, yeah, sure. Let's say that you worked for an organization that had very stringent information sourcing requirements. Maybe there was an internal database where you had to log the URLs, and the authority, and there was requirements that everything be authoritative. Our tool, because it's pointed at these PUC documents, can give you the underlying document, where it was sourced, the PUC regulatory authority. Not only that, it cites the individual page number, and if you click on the links, it'll drop you right into the spot that you're at. It really makes it so that this tool can be trusted to be using only information that is valid or authoritative.

We run into questions like there's this company and they do this web-crawling technique that's pretty cool. But if you're not allowed to use Reuters or another company's articles in your underlying research database, then that's no good. Then a lot of companies, if they're doing due diligence, they can't do the newspaper article themselves, they have to find the PUC article that's underneath the news article. We just cut out that middle man right there. That would be one of the various things.

The other one would just be, on the back end, we're able to do what I would call a boil the ocean approach. If you're interested in data centers, we can find every last word data center in every docket. We can find the word crypto, or miner, or any of those things, and retrieve every last docket so that we know that we have everything. Then be able to dig through them to find all of the pertinent information. Just knowing that you have your arms around this one big universe is important.

Then the other one, you mentioned why wouldn't you just use Google? It would be terrible if I woke up tomorrow and Google had crawled every PUC. But it was not easy. It took a team of crack engineers months and months to do it. I just don't see them being able to put forth something rapidly that's as good as what we have. I don't know if they want to. We're in this vertical AI space or verticalized AI. There's the foundation models that are at the bottom layer, and then there's all these applications that people build on top of it. That's much more of a wild west open ecosystem that we're living in. I think people like Google are really enjoying what we're doing because we have not only the AI tokens that we're using, but we're also storing in the Google Cloud platform, and we're using [inaudible 00:24:24] other back end stuff. I think that we are really good for that ecosystem in general.

Sam Humphrey:

I am curious about where this could go because you did mention it has applications outside of the US. You've now got all 50 states. But if you wanted to, say expand this into Canada, or I don't know, Mexico, how challenging is that? Is that a whole host of legal requirements to get that data within your models, or is that a possibility in the near future?

Alex Klaessig:

Well, Canada's absolutely on the roadmap for the near future. Several provinces are interconnected with US provinces. We share a lot of the same regulatory requirements, in that regard. Now that we've hit 50 states for electricity and natural gas, we have a couple of choices to make. One is going back in time. Right now, we only did 2020 onward. We're just going to turn back the clock and get 1990 onward, or whatever an interesting point seems like.

Then the other thing is that these PUCs regulate other utilities and they regulate other elements of the US economy. Things like broadband, telecoms, medical insurance. We started with electricity and natural gas because that's my personal subject matter expertise and Bruce's entrée into this particular world. But once we crack the energy utility, we'll move on to other utilities and we'll move on to other things. I think water would be the natural next step because it's very similar in delivering a commodity. That's just really from an interior engineering choice, we just have to make one change to the back end crawlers, and all of a sudden that new database will flow in.

Then the other choice is do we double-down on other sources of information in the United States? The North American economy is so information-rich. There's all sorts of state regulatory permit information that we could pull in, state architectural boards that have another rich layer of information that people are very interested in. We're letting our engineers take a breath and celebrate this, and maybe work on a couple of projects that aren't crawlers. Then we'll share the news with them that they got to get back at it.

Hill Vaden:

Personal question for you. We were lucky enough to work together and I know your history as a great energy analyst. What are some of the moments working at Halcyon where you've seen something come out of your project team and said, "Holy cow, this is exactly what I needed?" Or, "Oh my God, this totally would have changed how I was doing things." Have you had several or at least one of those eureka moments since moving into Halcyon?

Alex Klaessig:

Oh, yeah. I have the opportunity to work with the software engineers, they're really talented software engineers. I was talking to them one day and I mentioned how, a lot of times when I'm going through documents, is I'm just opening them up and skimming them to see if there's a cool chart or a cool graph in there, and then I'll read around that. They made the chart extractor. If you run a query, it will look around the answers that it found for all of the relevant charts and tables and it'll show them to you. Not only did I think it was such a neat little cool tool, they did it in a week. It just was I mentioned it, they come into the next team meeting, and there's this tool that allows me to find the tables, find the charts, and then I can click a button and it'll download them all into Excel for me. I was like, "Oh my Lord, this is exactly what I do so often when I'm an analyst."

It was a real whiz-bang moment where I was like, "This is so interesting and such a delight to be able to have this resource that I work with now."

Sam Humphrey:

That's so cool. Just looking at time, we always want to round these conversations off with what are some really key things that we, just the three of us, and the rest of the people listening should be looking out for with your company over the next six months to 12 months that you want us to pay particular attention to?

Alex Klaessig:

We just recently hit the 50 states available in our platform and that means that we have nationwide coverage for our alerts. It also means that we can start extracting the information. What we're really looking forward to over the next six months to a year is getting some data products out there and some data feeds to customers that give them access to previously inaccessible information, and then using that to raise our Series A and be able to grow the company a lot more.

Hill Vaden:

Before we hang up, just one question on customers. Is the customer an individual, or is the customer, is a B2B-type model?

Alex Klaessig:

Oh, B2B. You're the customer, Hill, I want you to be [inaudible 00:29:29].

Hill Vaden:

Okay. Well, we'll put information into the abstract about Halcyon. This has been a fun conversation and I really enjoyed the chance to reconnect. We look forward to talking again soon.