¶1Hello, hope you are doing well. So, I'm back with another video. Something I just wanted to talk a bit about because it's been working out so good for me. And this is maybe you can maybe look into yourself if you are kind of interested in this. So what I've been doing I guess over the last couple of weeks is I've been using AI mostly like codeex claw code open code to help me do a lot of research on kind of execution alpha I called it on polyarket.
¶2So basically this has not so much to do with the prediction itself. we rather try to optimize something that is uh more technical like how we actually execute our orders and stuff like that. So what it is uh is basically how fast we can kind of place an order on either on like a new position or on a existing event. And that is what I've been researching and trying to figure out how to get the best Q position. So this is kind of what could give us a small edge, right?
¶3So I'm going to talk a bit about how I do that, what I found out and is there any edge, right? Is there is is this profitable? You can see an example here of a 10,000% profit. So here you can see I guess I can show you here. So on the Bitcoin, this is a five minute up and down test I did.
¶4So I bought um five shares for five cents, I guess. Uh I guess that's and when this resolved, you can see we got $5 back here. Okay, so that's like a big profit. But of course, this doesn't happen every time, but you kind of need a Q position to actually get uh a good probability of landing these types of u orders, right? Or fills I can say.
¶5Uh another example I guess I can show you is if we go to my terminal. So here is another position I'm in. So this is US recession by the end of 2077. So that's a far out position. So this is just something I've been testing and this is like a very commitment but it doesn't really matter too much because uh you can kind of see here.
¶6So we have the legs yes or no and I ordered five shares at 1 cent. So this is of course resting now on this uh event. And here we can kind of see the execution. So we can see we have established or estimated shares ahead 12. Okay.
¶7But if you look at uh estimated shares behind us, it's 1.1 million, right? So this means that we are like 0 0% 1% ahead of all the other people in the queue. And if there is some kind of event that kind of spikes this market, we probably will get filled at 1 cent. Right? So if we head over to Poly Market and here you can kind of see it.
¶8So if we go into this event and we kind of scroll down here and look at the order book here. So you can see uh for yes you can see the order book down here where we have our position right five shares it's 1.1 million shares resting. So everyone is behind us and that gives us like a what kind of edge can do you call that? more of like um it's not like a predictive edge but it's more like a technical edge I could call it because this position is kind of worth more than being at the end of the tail of the order book here right because this uh on poly market we use something called we use something called uh first in first out or FIFO. So basically if you uh are ahead of the queue here you will be filled first right.
¶9So that is the kind of the execution alpha we try to uh yeah try to optimize for so let's say the market spikes now and like at a brief second uh the order book comes down to 1 cent we are already far ahead here. So we will get filled. Let's say only 100 people get filled. We will get filled at 1 cent. Uh but the other 1.1 million would not get filled.
¶10So that is kind of our edge. And there are some value in being this early uh on a order book. So this is something I've been trying to look more into lately and kind of how I can use this or how I can do this. And there's a lot of different things you can do. So I thought I could give you like a couple of tips how my execution has been improving so much uh by using research with AI here.
¶11So these tips are kind of generic but uh I think they are a good place to start if you want to try to start optimizing and kind of involving more uh aentic AI or something into your pipeline because uh if you use like codeex or cloud code um it's really good at kind of optimizing your code uh just based on like these simple instructions. So of course we want to use websockets that kind of is kind of given right we don't want to be polling uh so we always want to set up like a websocket with kind of yeah an open connection to the to poly market that is also pretty good and we can also kind of premputee a lot of parameters and credentials so you can always think of it like it's cache or something like that almost uh hosting is important right so depends where you are where your servers are. So, uh this can be done by using like a VPS or something. That is also something you should look into. I think that's uh you can earn a lot of uh micro milliseconds on that.
¶12Uh of course, reuse uh persistence uh HTTP connection connections is also working pretty good. And we have profile signing and execution latency. So basically pretty generic tips but these uh are just a few you can start with and just keep iterating testing and kind of running more tests research parameters. I think I spent a lot of days even trying to get mine working pretty good and I still have a lot of research to do to optimize this. But uh if you are kind of interested in stuff like this, this is a really uh good way to leverage AI because of the coding and the speed you can iterate and run experiments.
¶13So this is something I have a lot of fun with and I'm always running some kind of experiment to see if we can kind of get our milliseconds uh even down more right than we had before. And for me, I'll be kind of slowly creeping down. Of course, it took a lot of time and trial to kind of get down to being like 0 01%. Uh, and that is kind of random. It doesn't happen all the time.
¶14There's a lot of competition, but uh, sometimes you can get lucky and your FIFO position is very good. And this gives you some kind of more of like a mechanical edge, not of like a prediction edge. But, uh, still quite fun. But of course, you have to be interested in stuff like this. Not just only the outcome, but more of the technical side around it.
¶15So, kind of my plan going forward is to kind of look a bit more at these repetitive uh Bitcoin crypto markets and trying to kind of put my AI agent at looking at where we can kind of look in the order book to find some kind of more mechanical technical edge uh on this. So I don't know if it is possible but uh on my testing I had a few good incidents where we got that full onent fill and the full payout and of course the math behind that is uh very profitable because the draw down is so small and you can kind of ride many many times on the adversal side but still come out ahead if you have a lot of patience for that right so basically that's what I wanted to share today just a bit of my research into AI execution. Uh I also have been looking at this on Calshi and on Hyperlquid and other platforms what we can do there. Uh but that's just going to have to be another video. This is kind just kind of where I put most of my effort in now.
¶16And there we have seen some type of results uh like we did here with like a up and down market and also on the more general events market has been pretty interesting. So, if you want to more content like this, just give this video a like, subscribe, and yeah, hopefully I'll see you again on my next video where I will kind of do the recap of this week's prediction and some other ideas I'm working on, and I'm going to share a few results and stuff. So, yeah, thank you for tuning in. Have a great day and hopefully we speak