Igor Litvak So let's start off with a warmup question. Let’s take a look here. Top 20 Most active CUSIPs in Energy last week. So, let’s say you log in on a Monday, and you just want an overview of your sector from last week. And this is a great starter question because it helps with idea exploration. After you examine the bonds that come back, you might want to dig a bit deeper into a specific area. Let's say only the five year here. Now show me only the five year bonds.

 BondGPT allows you to ask follow up questions like this one so you can continue your exploration. So, there is an Excel export button here at the bottom, but in the same amount of time it would have taken you to open the Excel file and create your filters, BondGPT has already filtered the results for you down here. So, we really encourage everyone to continue asking follow up questions as they go down and ultimately whittle down the list to the exact bonds that you're looking to go through. Now, let's say you want to dig a bit deeper into the energy sector again. High yield upstream bonds with yields above 8%. So, let's say you have a specific need for yield and want to filter on that primarily. And that's what this question is looking for.

 In this question you'll notice some specific trader lingo in the term “upstream” which is unique obviously to the energy sector. As we were building out BondGPT, we ran into several situations where BondGPT needed to understand common bond language and common trader lingo. As another example, you can ask BondGPT information on beer bonds or boats, and it'll understand all of those as well. We've programed all those common terms into it. So, this entire exploratory process of finding bonds to trade is made much faster by BondGPT. So, you're not wasting vast amounts of time on querying because results are coming back so quickly. So, your iterative process is much faster. So rather than having to manually pull up different databases and play around with different user interfaces, you can just play around with different prompts just by typing them in. And you're not being penalized on time for your exploration.

 And now let's turn to the complexity a bit on BondGPT. Say I need some front end paper with enough yield to justify buying credit over Treasuries. I've got some sector barriers. Maybe I'm already overweight in financials. I want to contain risk, so I limit the price volatility. And then I want a relatively high issuance so I have a better chance of actually finding these bonds in the market. Let's run this question. Now, again, the key here is time efficiency. So imagine running this question manually. And maybe the first part of the question is fairly straightforward and wouldn't take too much time.

 But the second half that's asking about price volatility would easily turn this into an hour-long exercise. And that's assuming I don't get tied up with other tasks or I need to go and trade my blotter, right? So instead of all that BondGPT can run this in the background for you in a fraction of the time. Now let's look at what's happening here. So, the first thing we do here is we repeat the question to you in slightly different terms so you can be confident that BondGPT truly understood what you're looking for. We then took it a step further and created our patent-pending “Show your work” function. Now, again, we want you to see exactly how BondGPT is coming up with your answer so you can be confident in the results. We take you through, step by step, the databases that we're querying and how we're coming back with your list of bonds below. You can also see here that this query took 15 seconds versus an hour. And we're letting you know this was not a follow up question. Again, BondGPT understands that this was a new and unique question. We can also see here that 82 total bonds fit the criteria that you've typed in. And down here you can see the results. We include a bunch of additional information on top of the information that you asked for. So you see you're asking for price volatility. So we're giving you pricing, of course, issuance, so we're giving you amount outstanding, S&P rating tenor, so on and so forth. A lot of good information. Where you're going to see here is the Cloud Score. Now this is LTX Proprietary Liquidity Cloud Match Score. This is a great addition to your query because while we found a list of bonds that fit your criteria, your next question will be, Am I actually able to find these bonds in the market? Can I get liquidity? So this is on a scale of 1 to 10, ten being the most liquid. LTX’s Liquidity Cloud Score lets you know which bonds are currently active on the LTX trading platform and that will tie in nicely when we when we go back to talking about list trading using BondGPT, which we'll get to in a little bit. Now let's look at a query that might interest traders in the ETF create redeem space.

While this is running, let me reiterate a point that Fitim made earlier and that we often get asked. So people always wonder, can BondGPT produce results as effectively as a programing language like Python or R? And the answer is that BondGPT doesn't compete with Python, but rather it's an evolution in how users can access data. So data itself and how we analyze it is becoming more and more complex every single day. And that means less people can readily access it. BondGPT is a huge leap forward in allowing any user to quickly access vast amounts of information in just seconds, as you've seen, without needing to know special programing language. So BondGPT doesn't compete with the data tools that are out there. But it really it brings the strength of those tools to a greater number of people.

So let's look at this list of bonds. So a few minutes ago during the PowerPoint, we were talking about GenAI powered list trading. Any output from BondGPT can lead directly to trading on LTX. When you're working with BondGPT on the LTX trading platform and you find your list of bonds that fit your criteria like we see here, you're able to create an inquiry list directly from the results and then quickly send out that list for pricing on LTX.

Now let's go back and talk about relative value and see how BondGPT queries LTX’s newest model. So we're going to ask: Show me relative value opportunities in the ten year. So let's talk, about while this is running, we'll talk about the GPT acronym. It stands for Generative Pre-trained Transformer.

 As we mentioned before, the generated portion of BondGPT has a temperature that we have essentially turned down to zero. Now let's take a look at these results. So again, as we talked about earlier, for relative value, what we're showing you are potential trading pairs where one bond is cheap and the other is rich in relation to one another. So, we have here CUSIP one and CUSIP two, and we're not making a recommendation of which one to buy and which one to sell. Rather, we're letting you know that one is cheap and one is rich in relation to one another. And what we're providing you here is a purely quantitative result. Again, we've looked at establishing correlation and then we're looking at divergence. And that's what we’re providing you here is the pure quantitative result. And then you as a trader or a PM research analyst, you are then providing the qualitative layer of the analysis on top. And if you remember, we're using an ensemble of six cutting edge relative value models to generate these answers. And through our regression testing, we've further found that requiring four out of those six models to agree on a recommendation ended up producing the most reliable results. So in other words, the results you see here have passed the test of at least four of the six models that we're using. So that means that the relative value recommendations are very conservative by design. So we've seen a lot of interest from the sell side for this model as well since coming out with pair trades is something a sell side trader is expected to do manually for their specific sector, whereas BondGPT runs the analysis on roughly 30,000 CUSIPs.

Now let's run one more that's querying multiple databases in one go. So here I want to highlight the portion of this question that's looking at revenue. And emphasize again what Fitim said about BondGPT being able to process structured data, meaning the information is in a table as well as unstructured data such as reading a PDF, which is exactly what we're seeing here. So BondGPT is reading through 10Ks, 10Qs, to return back the information and is reading it from a PDF.