WEBVTT 00:00.480 --> 00:06.400 And so this is the moment now when we make the transition from being a scientist to an engineer. 00:06.440 --> 00:08.040 We've done our experimentation. 00:08.040 --> 00:10.640 We've explored some agent configurations. 00:10.640 --> 00:18.160 We've we've figured out that this combination of researcher and the trader makes sense and is working 00:18.160 --> 00:19.120 well for us. 00:19.240 --> 00:23.240 And now we're going to move to building our Python module. 00:23.360 --> 00:24.480 So I want to take you there. 00:24.480 --> 00:27.520 It's called traders Pi and it is quite long. 00:27.600 --> 00:28.440 There's quite a lot here. 00:28.440 --> 00:34.200 But it is just everything we've just done in that lab turned into a module. 00:34.360 --> 00:36.720 There's, uh, there's nothing else to it. 00:36.720 --> 00:38.440 So you already know this stuff. 00:38.640 --> 00:41.440 Um, it's just presented differently. 00:41.640 --> 00:45.560 Uh, and it's one of the reasons, again, it would be a lot to take on board to go through all this 00:45.560 --> 00:46.520 code for nothing. 00:46.520 --> 00:50.880 But if you've built up to it in a lab, then you're ready to construct this. 00:51.360 --> 00:54.080 So this is this is the there's imports. 00:54.080 --> 00:55.480 There's load dot env at the top. 00:55.480 --> 00:56.080 Of course. 00:56.200 --> 01:00.040 Now this here is one thing I need to draw your attention to, because it's the one thing that you might 01:00.040 --> 01:01.320 need to come in and change. 01:01.520 --> 01:08.600 If you're using the paid market data MCP server, then you need to turn this to being your directory 01:08.600 --> 01:09.560 where that's running. 01:09.720 --> 01:13.040 And sorry to give you something to come in and do in the middle of this code. 01:13.040 --> 01:15.880 But but that's just the one thing that is coded there. 01:15.880 --> 01:21.720 And I don't know where you will have git cloned that repo, so I can't speculate on that for you. 01:21.720 --> 01:23.400 So I'll let I'll let you do that. 01:23.720 --> 01:28.640 Um, and then basically this makes a determination whether you're going free or not based on if you've 01:28.640 --> 01:34.480 got an API key for the for the paid product in your EMV, then it assumes you want to use that. 01:34.480 --> 01:37.680 So only set that API key if you actually want it to use it. 01:37.760 --> 01:40.720 Uh, and again, you know, it's it's $0.01 for every lookup. 01:40.720 --> 01:43.400 So just just be aware of that. 01:44.040 --> 01:47.360 So we're also going to get the deep seq API key and the Google API key. 01:47.360 --> 01:53.640 Because in time we're going to be using a combination of open AI and deep SEQ and Gemini together to 01:53.680 --> 01:54.400 do this. 01:54.400 --> 01:55.800 But you don't need to do that. 01:55.840 --> 01:58.320 And for now it's we're not we're not going to touch this today. 01:58.320 --> 01:59.400 That will be for tomorrow. 01:59.560 --> 02:03.040 Um, and I've got the base URLs coded here so that we can use it. 02:03.360 --> 02:05.520 And Max turns is something I'm using. 02:05.720 --> 02:10.960 Uh, you may remember this from, from when we used landgraaf that Landgraf would, would error out 02:10.960 --> 02:13.150 if it took more than 25 turns. 02:13.150 --> 02:14.950 So this is something to control. 02:15.270 --> 02:15.550 How? 02:15.710 --> 02:20.110 How deep down, the agent interactions will go before it will throw an error and say. 02:20.150 --> 02:21.350 Okay, enough is enough. 02:21.950 --> 02:22.590 All right. 02:23.470 --> 02:30.470 So the way with OpenAI agents SDK that you specify, if you want to use a different client to swap out 02:30.470 --> 02:37.230 from OpenAI is this way you create an async OpenAI instance like this. 02:37.270 --> 02:41.630 And so this is showing the Gemini client as an example. 02:41.870 --> 02:52.030 And we will uh, use that potentially in this case for, for uh, Gemini Flash right there. 02:52.630 --> 02:54.790 And somewhere there should be a deep seat client too. 02:54.790 --> 02:55.550 I don't see that there. 02:55.550 --> 02:56.990 So we should put that in right now. 02:57.470 --> 03:00.470 Let's copy this and make a deep neck client as well. 03:02.190 --> 03:03.270 Deep seek. 03:05.630 --> 03:06.390 Client. 03:07.710 --> 03:11.150 Uh, and this should be, of course, the deep seek base URL. 03:11.390 --> 03:14.190 And that should be at cursor has helped us out there. 03:14.190 --> 03:15.270 That is perfect. 03:15.310 --> 03:16.110 Okay, good. 03:16.110 --> 03:17.670 We now have our clients. 03:18.510 --> 03:19.150 All right. 03:19.150 --> 03:26.070 And then this next part right here is simply setting up the MCP servers, just as we did in the lab. 03:26.390 --> 03:28.990 Uh, the trader MCP server parameters. 03:28.990 --> 03:34.950 It's going to use accounts and it's going to use memory if we're using free market data. 03:34.950 --> 03:40.310 If we're not if we're using paid market data, then we're also going to add in the paid MCP server. 03:40.830 --> 03:47.390 And otherwise if we're using the free one then we will just we will collect the tools later. 03:48.070 --> 03:55.750 And then for the researcher we're going to use the MCP server for fetch and the MCP server for brave. 03:56.150 --> 03:58.110 And that is set up our params. 03:58.110 --> 04:04.790 We've now organized the params for our MCP servers into different lists, and we're ready to get going 04:04.790 --> 04:05.670 with the code. 04:05.710 --> 04:07.590 All right, so now we've set up our params. 04:07.590 --> 04:12.110 We now have three utility functions before we get to the meat of the module. 04:12.310 --> 04:14.590 First of all getmodel very simple. 04:14.590 --> 04:21.590 You pass in a model name and it's going to return one of these OpenAI chat completions models or string 04:21.700 --> 04:23.700 if it's if it's just GPT four or mini. 04:23.820 --> 04:29.460 So so if you pass in deep chat, it's going to create one of these guys using the deep sea client. 04:29.860 --> 04:33.620 If it's Gemini, it will create one of them using the Gemini client. 04:33.660 --> 04:37.100 Otherwise just the model name just to simplify some of our code later. 04:37.660 --> 04:39.980 And then this is Get researcher. 04:39.980 --> 04:42.020 This is exactly the same as we had before. 04:42.060 --> 04:44.860 This gets our researcher agent. 04:45.060 --> 04:51.060 It's got the system prompt for the researcher uh including the current date just as we said. 04:51.300 --> 04:57.180 And then it creates the agent and we pass in the MCP servers, just as we did before. 04:57.220 --> 05:00.540 We populate them and we return the researcher. 05:00.740 --> 05:08.660 And then the final utility is to wrap this in a tool, return the researcher as a tool, because we're 05:08.660 --> 05:11.660 going to use it in the request response kind of mode. 05:11.980 --> 05:18.900 Um, and so for this we this just calls the get researcher function and then returns that as tool. 05:19.100 --> 05:23.940 As tool being the way that you convert, uh, the, the agent into a tool. 05:24.420 --> 05:29.450 And we give it a name researcher and we give the description of the tool just as we did before. 05:29.690 --> 05:32.450 And that is the end of our utilities. 05:32.450 --> 05:35.770 We're in to creating our class trader. 05:35.810 --> 05:37.250 Okay, here we go. 05:37.250 --> 05:38.810 So a trader has a name. 05:39.210 --> 05:40.970 It has an underlying agent. 05:41.010 --> 05:43.050 It has a model name we'll set it up with. 05:43.330 --> 05:49.770 And then it has something that that it has an account that will populate when we read its account details. 05:49.970 --> 05:53.090 We store its MCP servers right here. 05:53.090 --> 05:56.770 And then a combination of the MCP servers of both. 05:56.770 --> 06:03.610 So this here we're actually instantiating, we're creating an MCP server studio instance for each of 06:03.610 --> 06:07.010 the different parameters uh, by iterating through them. 06:07.010 --> 06:10.890 So they are held at the instance of trader okay. 06:11.490 --> 06:21.210 And then we have here something which is going to, uh, set the system prompt for our trader agent 06:21.410 --> 06:25.810 and it says you are its name, a trader on the stock market. 06:25.810 --> 06:27.490 Your account is under your name. 06:27.730 --> 06:34.170 And now this this stuff is basically the same as what we had in the lab with one tiny difference, which 06:34.170 --> 06:34.890 is that I'm. 06:34.890 --> 06:39.850 I'm saying that you actively manage your portfolio according to your strategy, but I'm not actually 06:39.850 --> 06:43.690 getting it to put its strategy in the system prompt. 06:43.930 --> 06:45.130 And so it's not going to do that. 06:45.130 --> 06:47.090 It's going to put that in the user prompt. 06:47.290 --> 06:49.530 And why is that you may ask. 06:49.770 --> 06:56.970 Well that's because I have wild plans in the next uh, time in the in tomorrow, we might update this 06:56.970 --> 07:02.410 so that it can actually change its strategy as part of what it does, because I think that would be 07:02.410 --> 07:06.890 really cool if it has the ability to actually change its own strategy. 07:07.130 --> 07:09.970 Uh, we'll, we'll see if, uh, if we want to do that. 07:10.290 --> 07:12.330 Um, but, um, yeah. 07:12.330 --> 07:20.770 So, so, um, there's then get tools, which is simply going to, uh, populate the tools that this 07:20.770 --> 07:21.210 has. 07:21.450 --> 07:28.970 And if you're using free market data, then this is where it's going to to, uh, use the MCP client 07:28.970 --> 07:35.170 that I wrote and then use that to build some tools that we can call directly from our LM. 07:35.850 --> 07:41.930 The Researcher is getting that, using that utility function to get a tool that wraps our researcher 07:41.970 --> 07:42.570 agent. 07:42.970 --> 07:45.090 And then we return those tools. 07:46.290 --> 07:52.970 This this function connect to MCP servers iterates through all of our MCP servers and calls connect. 07:53.890 --> 07:54.650 All right. 07:54.690 --> 07:57.250 And then init agent does those. 07:57.290 --> 07:58.650 The connection I just said. 07:59.170 --> 08:02.850 And then it gets the instructions gets the tools. 08:02.850 --> 08:08.210 And then it creates the underlying agent that it stores in the in the instance variable agent. 08:08.410 --> 08:10.170 And it is an agent with its name. 08:10.490 --> 08:15.410 It could be Alice, it could be whatever with its instructions with the underlying model. 08:15.650 --> 08:23.010 And this is where it calls that utility function to, to to uh to get model given its name, the tools 08:23.210 --> 08:26.370 and the MCP servers are all set up. 08:26.370 --> 08:32.850 So now the underlying agent instance variable will now be populated with an agent. 08:33.570 --> 08:39.010 And it gets the account to the starting position and it returns the agent. 08:39.050 --> 08:39.690 Okay. 08:40.330 --> 08:41.250 So we're almost there. 08:41.250 --> 08:43.130 We're almost ready for action.