¶1Hello, hope you're all doing well. So, today I just wanted to do something fun. I wanted to see how fast we can actually go from just starting with nothing in QuantX to actually run all of these kind of set skills here. So, of course we need some data, but let's say on QuantX we can kind of get the data from some historical stuff. And I wanted to actually check out the BTC 15-minute market just to see what data we can find there.
¶2And when we kind of have the data or at least set up the data so we can fetch it, we we're just going to run my skill here that I kind of have a pre-organic used for kind of finding hypotheses that we can research. This is like an optimized skill.md file I have. And I also have my hypothesis tester. This is more like doing some backtesting, walk forward, out-of-sample, and stuff like that with the hypothesis we create. So, I thought it'd just be fun to just run your tool like a speed run and see how fast we can actually get something up and running.
¶3And when we kind of have tested our skill, let's see if it's somewhat interesting. We can just use QuantX GPT 5.6 today, I think, uh to write the code for this and then we can execute it on QuantX. So, will we make any dollars over here? We don't know, but I just wanted to see how fast we actually can do this kind of like a speed run, like I said. So, yeah, I think we're just going to get started and see if we go from like nothing here to fast we can get running.
¶4So, if we go to uh QuantX, you can see we have something called historical data. So, I think we're going to just try to use that. So, let's just copy that. And yeah, I'm going to use QuantX. Uh let's do YOLO.
¶5Yeah, I think that's what we're going to do. Okay, uh I think I'm just going to select on high and that's fine. So, let's just do read uh the historical data. So, now we have that and we can check our skills. So, list you can see I think I have like a quant research and something like uh quant hypothesis research or some skill something I set up uh a while back.
¶6So, you can see now we kind of have uh looked at the data here in Codex. That should be fine and let's just uh do a prompt here. Okay, so I'm just going to do uh we want to target the Bitcoin up and down 15-minute market. You know where you can get get the data. We just linked that, okay?
¶7So, start finding uh some hypothesis. Yeah, I wrote this wrong. Uh we can test to find an interesting strategy to run on this event and I'm just going to tag my skill here. So, this is just a research skill. It's nothing like special, but it's just like a very structured way to do this.
¶8Uh and it kind of yeah, says don't do any testing. Just look at the data, ask some questions around what is strange with this data and stuff like that. And let's see if we can find anything interesting here. So, yeah. Um let's see what happens.
¶9Okay, so you can see now we're just going to do I'm using the skill, right? And now it's going to go straight to looking at the data we kind of want to fetch for this. Let's just see. Yeah, it's looking at the field timestamps. It calls historical coverage.
¶10All of that looks pretty good. So, you can see uh Codex creates a plan here. It's going to look at the um BTC artifacts data sets. Verify this and it's just going to start looking for hypothesis. So, now we're just going to let this skill run out and when we come back, hopefully we have a few different hypotheses we can test and run the other skill we have created for this before we actually hopefully find something and then we take it online.
¶11Okay, so unfortunately, I think I lost some of my, yeah, recordings. I just lost it. I couldn't find it anywhere. But basically, what we did is when we got those five, uh, actually, from the prompt, I got five different hypotheses that we could actually run some testing on. And I did run I'm going to show you the prompt I did run to actually test this.
¶12Like I said, it was another pre-made skill. Uh, I might try to put them up somewhere to show you where the what kind of skills I'm using. Uh, we found five different ones, right? And we ran them through. Uh, we didn't find any like great ones, but we did find someone that is we can keep researching on.
¶13Two were rejected. This was like some, yeah, outcome alternation 5-minute, minute five under dispersion. So, we have three, and I asked what was the promising one. That was number one. That was like some volatility persistence.
¶14And I wanted to know estimated trades per day because we can just kind of look in the data. Uh, so we have a top third volatility filter. That's like 32 trades a day. Uh, I was thinking maybe we can set this up just to see. I might just keep this running overnight.
¶15And tomorrow morning, I will publish this. And let's just see uh, how it does. So, I think we're going to try that. So, let's check out the state of our Kalshi account now. So, yeah, like you can see, we on this account, we have like 120 bucks.
¶16That's fine. I think that's fine just to test this out. So, basically, what I want to do now is I'm just going to ask Codex here to help us kind of write the code for this. Uh, I also went ahead and got my, yeah, credentials to access uh, the API for Kalshi. So, I'm just going to keep it simple.
¶17Uh, let's just go for a top third 90% filter with 32 trades per day. So, write this code, uh, to execute the The precisely and according to the spec, cred uh, credentials to culture account is in the EMB. Test, let me know when you are ready or we are ready to run this. So, let's just wait for code x here 5.56 all. I had this on high to write a code for this setup, right?
¶18We got a small plan, execute the code, do some testing, and pretty much we should be live starting to putting down some orders and kind of monitoring this. I might just say uh use web sockets. Uh instead of poll. I guess that was the plan, but I just in case because we don't want to use polling. We always want to use the web socket, right?
¶19So, yeah. I'm just going to let this run and I take you back when hopefully we have something that is actually ready to to launch. Okay, so you can see now we are up and running here. So, that should be pretty straightforward. So, uh it doesn't look like we have any positions yet.
¶20Yeah, those These are some other tests I'm running. Let's just ask a bit about the filter. Okay, so let's take a look at the filter here. So, here is kind of the strategy, I guess. So, look at the immediately previous 15-minute contract.
¶21Calculate its absolute BRTI move from start to close. You can see it must move at least 15.4 basis points placing it in historical top tier of for uh for volatility. Uh in the current contract at 5-minute close, calculate the yes midpoint. So, if it's above 50 cent at this point, we can do yes, it's be- below, no, exactly 50, skip. Pretty straightforward, right?
¶22Then wait one complete minute and at minute six, place the $5 fill or kill order using a fresh web socket book. So, the timeline is previous close, the volatility filter passes, current 5 minute, choose favorite at 6 minutes, submit the trade trade if the depth and timing check pass. Pretty straight strategy. So, I think that could be a pretty interesting and we should be up and running here now. So, what I'm going to do, it's getting pretty late here now, so I'm just going to leave this up for the night.
¶23And when I get up this morning, I'm going to see how we're doing and I'm going to take you back. Let's just ask running healthy before we before we end this for tonight. Okay, yeah, we are running healthy and you can see the current window will skip the preceding BTC move was only four basis points be- because we need 15.4 for the volatility, right? Uh so, yeah, we're just going to wait until the morning, I think, and let this this run overnight. Uh I want to tell you a bit about some new projects I have.
¶24This is going to be kind of a micro hedge fund for AI agents. Um I will talk a bit more about it, but basically uh I want to try to create some kind of small hedge fund. I would just want to call it that. And to try to run all the strategies I've been running on prediction markets with some really good percentage results, at least. I haven't really put much stake behind it yet.
¶25But you can see if you go to max with two x's quant.com, you can kind of see some of my strategies I've been running here. Uh and you can see they don't really lose. I guess we spent some research uh money in the beginning on this account, but after that, yeah, we kind of go pretty straight. It's not like straight up, but uh we don't really lose that much. We have some drawdown, but not too bad.
¶26So, that is something I'm working on, so I'm going to come back to that, but the basically, if you are interested, uh just check it out. And the idea is that you can um So, you can kind of start at 1 millionth of a dollar. There's no like you can invest like basically nothing for AI agents uh on the crypto base or something. I think I'm going to do. So, yeah.
¶27That is something I have coming up. Uh just a fun project. So, I'm going to talk a bit more about that later. But you can read the description and read a bit about it if you want to. So, yeah.
¶28Uh for you, I'll be back in 1 second. It's going to be tomorrow morning, I guess. Uh let's see how it went. Okay, so it's the next night. Uh I just stopped the bot, but you can see here we did pretty good.
¶29So, now our balance it at 137. Uh over the last 24 hours, we had a 18 uh almost $20 up with a 15.7% gain. So, yeah. Our model did pretty well overnight. Uh not model, our strategy.
¶30So, remember this was a super simple strategy. We just had some set point. We did that. We did that. We did that.
¶31You remember the strategy, I guess. So, at this point in time, at this market, this went pretty good. But uh like I said, we saw in the data, I don't have like a big um uh like a expectation of good returns over long time for this model. But in periods, this could be super interesting. And with this video, I just wanted to show you that if you kind of want to get into some more systematic trading, some more autonomous, so maybe some AI-infected trading, uh there are a lot of opportunities out there now if you want to use uh like uh GPT-5.6, Claude code, open code, Kimikaitree, GLM, other the all of these models, Quen models, to just help you set this up.
¶32There's some really stuff really stuff really fun stuff you can do now to actually learn more and test things so fast, hypothesis like we did here with some simple skills. I might try to leave those skills up so you can check them in the description. And let's take a look at our drawdown here. So, drawdown is how much we we went down before we kind of won. So, I think our biggest drawdown was $12.
¶33So, that's like three three losses or something. PNL was plus 15, something like that. 14 wins and four losses over 18 trades. So, pretty good win rate, but win rate, yeah. I guess in this case win rate is pretty important.
¶34But, yeah. 14 wins, four losses, not bad. So, yeah. I think this was just a super fun thing. Fun project.
¶35And this is something you can try out for yourself. So, yeah. Think that's what I wanted to talk about. I guess I mentioned my micro hedge fund for AI agents that I'm starting. Check that out.
¶36If you want to support the channel, I have some links to like Caltech, Polymarket if you want to support the channel. It's not sponsored, but if you want to yeah, support this. I think I will be back tomorrow with a new video. I'm going to do some cross arbitrage on Yeah, you can see it here. It's on cross arbitrage on Polymarket and Caltech with some AI setup here.
¶37So, that's going to be probably tomorrow's video and see if we can find some arbitrage between those platforms. So, yeah. Enjoy your Saturday, I guess. And we speaks hopefully very soon.