WEBVTT 00:00.240 --> 00:03.440 Okay, so let's close the terminal. 00:03.680 --> 00:05.480 Come into financial researcher. 00:05.680 --> 00:06.600 You know the deal. 00:06.600 --> 00:08.760 Now there is a source folder. 00:08.760 --> 00:10.120 It has a config. 00:10.520 --> 00:12.040 And that is where we want to begin. 00:12.040 --> 00:15.600 We want to begin by looking at our agents. 00:15.800 --> 00:17.440 So this is the agents. 00:17.440 --> 00:21.440 And we are in fact also we're going to have a researcher agent. 00:21.640 --> 00:27.760 And we're going to make this a bit shorter and just have an analyst agent, a researcher agent and an 00:27.800 --> 00:28.840 analyst agent. 00:28.840 --> 00:30.160 So what are they going to do? 00:30.440 --> 00:37.880 So this time I'm going to to simply copy and paste a whole new section here and talk it through with 00:37.880 --> 00:38.360 you. 00:38.360 --> 00:43.760 So the role is going to be a senior financial researcher for a company that we will specify in the inputs 00:43.800 --> 00:49.640 like before, it's going to research that company news and potential for that company. 00:50.040 --> 00:50.840 The backstory. 00:50.880 --> 00:55.920 You're a seasoned financial researcher with a talent for finding the most relevant information about 00:55.920 --> 01:02.090 a company known for your ability to find the most relevant information, present it in a clear and concise 01:02.130 --> 01:02.850 manner. 01:02.850 --> 01:07.290 And that's where we specify the LLM, which we can we can mix this up if we wish. 01:07.730 --> 01:13.330 And we now choose to have our second second agent, the analyst. 01:13.330 --> 01:15.330 Let me put in what I've got here. 01:15.770 --> 01:19.730 Market analyst and report writer focused on a company. 01:20.010 --> 01:24.650 You analyze the company and you create a comprehensive, well structured report that represents insights 01:24.650 --> 01:26.290 in a clear and engaging way. 01:26.930 --> 01:32.810 Meticulous, skilled analyst with a background in financial analysis and company research, a talent, 01:32.850 --> 01:40.290 etc., etc. it's interesting that this construct of having us tell the back story, it really does encourage 01:40.290 --> 01:41.330 better prompting. 01:41.330 --> 01:46.730 If I were just writing a system prompt instructions as we did last week, I probably wouldn't think 01:46.730 --> 01:53.090 through expressing myself this way in terms of using a word like meticulous and meaningful insights 01:53.090 --> 01:53.850 and so on. 01:53.930 --> 01:55.290 So it is helpful. 01:55.330 --> 02:00.870 Putting us through the discipline of having to give a backstory, helps give more context, and helps 02:00.870 --> 02:06.670 make sure that we're going to get the best outputs, the best outcomes from running the LM. 02:07.470 --> 02:08.470 All right. 02:08.670 --> 02:13.310 So with this, let's now talk about our tasks. 02:14.470 --> 02:15.150 Okay. 02:15.270 --> 02:17.510 So we're moving over to tasks. 02:17.510 --> 02:23.190 And we are going to have a research task and an analysis task. 02:23.310 --> 02:26.190 So it's not too different to the the boilerplate one. 02:26.190 --> 02:28.630 But we are going to do a little bit more work here. 02:28.830 --> 02:33.830 So for the research task going to really spell out what it is that we want this task to involve. 02:34.110 --> 02:37.190 Conduct thorough research on this company. 02:37.230 --> 02:43.510 Focus on the current status and health, the historical performance challenges and opportunities, recent 02:43.550 --> 02:46.710 news, future outlook, potential development. 02:47.150 --> 02:52.590 Make sure to organize your findings in a structured format with clear sections, and then the expected 02:52.590 --> 02:57.920 output is a bit repetitive, but it says what we want the well-defined sections and we again mention 02:57.920 --> 02:59.120 the input the company. 02:59.680 --> 03:03.040 And that is going to be assigned to the researcher. 03:03.560 --> 03:04.120 Okay. 03:04.320 --> 03:06.120 And the analysis task. 03:06.160 --> 03:07.880 Now you'll notice there's no output file there. 03:07.880 --> 03:09.960 So you might be thinking what what's going on here. 03:10.080 --> 03:10.640 Well let's see. 03:10.680 --> 03:15.480 Now let's give this a description that is going to be quite detailed. 03:15.520 --> 03:21.600 Again analyze the research findings and create a comprehensive report. 03:22.040 --> 03:23.560 The report should. 03:23.600 --> 03:25.440 And then we have the different sections in the report. 03:25.480 --> 03:32.120 Executive summary information insights market outlook and be formatted in a professional style. 03:32.560 --> 03:36.400 And for the expected output will again be a bit repetitive here. 03:36.600 --> 03:39.280 There's never a harm in being repetitive. 03:39.840 --> 03:48.160 Polished professional report on the company representing the research findings and the agent that we 03:48.160 --> 03:49.720 want to mention here. 03:49.760 --> 03:52.320 Of course should be the research. 03:53.040 --> 03:55.400 This is the the sorry the agent is the analyst. 03:55.530 --> 03:57.810 That's the agent that we that we made for the analysis task. 03:57.850 --> 03:58.370 Of course. 03:59.170 --> 03:59.650 Okay. 03:59.690 --> 04:02.210 So there's a few more things that we want to add here. 04:02.570 --> 04:11.450 We want to make sure that the output from the research task is included as part of the analysis task. 04:11.650 --> 04:14.330 And crew gives us a super easy way to do that. 04:14.370 --> 04:16.330 You just type context. 04:16.610 --> 04:21.090 And yes, of course cursor tells us what to do by saying that context. 04:21.090 --> 04:23.210 Context can take a list like this. 04:23.210 --> 04:28.810 And we are giving the research task as part of the context that's needed for the analysis task. 04:28.890 --> 04:35.450 And just with that simple step, we ensure continuity and we ensure that the output is included in the 04:35.450 --> 04:36.530 research task. 04:36.890 --> 04:40.250 And then finally we are going to have an output file for this. 04:40.570 --> 04:46.330 And we're going to let's let's call that report.md in the output directory. 04:46.330 --> 04:48.050 That seems fine. 04:48.770 --> 04:49.290 Okay. 04:49.330 --> 04:52.330 And now to the all important crew dot I. 04:52.530 --> 04:55.510 And here we have as usual, some gunk. 04:55.670 --> 04:58.270 We have the financial researcher I am. 04:58.270 --> 04:59.430 I'm going to be bold here. 04:59.430 --> 05:01.790 I'm obviously I'm going to leave in oops. 05:02.190 --> 05:03.270 Say that and I delete it. 05:03.670 --> 05:07.950 I'm going to leave in the config file that we want. 05:08.190 --> 05:10.990 And I am just going to delete everything else. 05:10.990 --> 05:12.510 And then we're just going to rewrite it. 05:12.510 --> 05:13.470 You want me to rewrite. 05:13.470 --> 05:17.150 Well we're going to rewrite this entire file right now okay. 05:17.190 --> 05:18.350 So what do we want here. 05:18.350 --> 05:19.710 We want an agent. 05:19.910 --> 05:22.230 So we put in the decorator for agent. 05:22.550 --> 05:25.070 And uh let's see. 05:25.110 --> 05:26.950 Well let's just press tab and see what we get. 05:26.990 --> 05:29.150 We want an agent that is the researcher. 05:29.150 --> 05:29.950 That's correct. 05:29.950 --> 05:32.510 We probably want to say verbose is true. 05:32.830 --> 05:33.630 There we go. 05:33.950 --> 05:38.070 And then we also want an agent that's called the analyst. 05:38.510 --> 05:45.910 And that is going to be an agent that is going to what's happening all over the place. 05:45.910 --> 05:46.190 Sorry. 05:46.190 --> 05:50.830 This is an agent and this should be an analyst. 05:51.630 --> 05:54.080 And now we're getting help from cursor. 05:54.080 --> 05:55.360 That all looks great. 05:55.560 --> 05:58.480 Okay, so now we want to have a couple of tasks. 05:58.480 --> 06:02.200 So we want to have a task that is called the research task. 06:02.200 --> 06:03.120 That sounds right. 06:03.120 --> 06:07.920 It's going to be a task that is going to the research task. 06:07.920 --> 06:08.920 That's perfect. 06:09.680 --> 06:13.560 And the analysis task that looks good to me as well. 06:13.920 --> 06:15.960 So all is well. 06:16.280 --> 06:19.920 And now we want the crew and yes thank you. 06:19.960 --> 06:21.320 Very nice cursor. 06:21.320 --> 06:22.960 It's filling it all in for us. 06:22.960 --> 06:31.240 We want the process to be the process sequential and verbose to be true. 06:31.440 --> 06:33.480 Crew cursor is faster than me. 06:33.520 --> 06:35.520 I'm playing catch up here but that is perfect. 06:35.520 --> 06:38.520 So remember self dot agent is populated. 06:38.560 --> 06:38.720 Sorry. 06:38.760 --> 06:46.600 Self dot agents is populated because we have agent decorators around our agent functions methods and 06:46.640 --> 06:50.200 then tasks is also decorated here. 06:50.200 --> 06:53.260 So everything looks great to me. 06:54.140 --> 07:01.900 Okay, now we go to Maine, and I think, again, we're going to want to, uh, delete all of this and 07:01.900 --> 07:02.500 start again. 07:02.500 --> 07:04.620 And I will type it for you manually. 07:05.300 --> 07:05.940 Okay. 07:05.940 --> 07:06.820 Here we go. 07:07.020 --> 07:09.540 So we're going to delete the gunk. 07:09.900 --> 07:14.020 We are going to delete everything but just only leave in run. 07:14.020 --> 07:15.300 That's going to be it. 07:15.340 --> 07:15.860 Okay. 07:15.900 --> 07:19.460 So run the researcher crew. 07:19.820 --> 07:26.660 Isn't it amazing that cursor realized knows that our input is a company that that inputs. 07:26.700 --> 07:31.420 We did indeed just have company as the templated curly braces thing. 07:31.460 --> 07:32.620 Just in case you don't know what I mean. 07:32.620 --> 07:34.300 I mean, we have this right here. 07:34.340 --> 07:38.660 Company is the one input that needs to go in to our YAML. 07:38.940 --> 07:46.260 And if we go back to main, then automatically cursor realized that we needed to set company and populated 07:46.260 --> 07:46.780 it there. 07:47.140 --> 07:47.660 All right. 07:47.660 --> 07:53.190 Then we're going to say yeah Result is the financial researcher. 07:55.790 --> 07:57.910 And that all looks great to me. 07:57.910 --> 07:59.830 And then we're going to print the results. 08:00.310 --> 08:00.790 Thank you. 08:00.830 --> 08:01.430 Cursor. 08:01.630 --> 08:03.550 And that's really it. 08:04.070 --> 08:05.670 I think that's all we need. 08:05.910 --> 08:08.310 Now you might be thinking hey, didn't we set up Serpa? 08:08.310 --> 08:09.630 Are we not going to use that for something? 08:09.670 --> 08:11.150 We are going to use it for something. 08:11.150 --> 08:12.550 But all in good time. 08:12.590 --> 08:18.310 Let's first just try running this without making any any proper Google lookup and see what happens. 08:19.190 --> 08:22.790 Okay, well, just before we run it, let's let's change the models. 08:22.790 --> 08:27.710 Let's give ourselves some more interesting models to experiment with for our researcher. 08:27.750 --> 08:29.590 Let's go with deep seek. 08:30.150 --> 08:32.870 Now of course you could do whatever models you would like here. 08:32.870 --> 08:33.590 This should be deep. 08:33.630 --> 08:34.230 Seek chat. 08:34.230 --> 08:35.910 That'll be the right one for us to use. 08:36.230 --> 08:38.350 And you could stick with OpenAI. 08:38.390 --> 08:42.310 Or you could use a different model, or use a llama and run it locally if you wish. 08:42.630 --> 08:46.550 I will try a grok model for this one. 08:46.990 --> 08:50.400 Let's use this this here llama three 370 billion. 08:50.440 --> 08:51.160 Versatile. 08:51.160 --> 08:55.440 That's one of the most powerful open source models from from meta. 08:55.640 --> 08:58.840 And we'll use that via grok the high performance inference. 08:58.840 --> 09:05.720 You can also run something locally like use a llama to to run llama 3.21 million or 3 billion version 09:05.720 --> 09:06.320 of it. 09:06.760 --> 09:09.920 So that's just a mix up the models to try something a bit different. 09:09.920 --> 09:15.440 And now let's go and open up a new one of these guys a new terminal. 09:15.440 --> 09:16.960 Let's make it nice and big. 09:17.440 --> 09:22.680 And we will go into project or folder number three for crew. 09:22.840 --> 09:25.920 We will go into our financial researcher. 09:25.920 --> 09:31.920 And you remember we type crew I run to kick this off and we'll see what happens. 09:32.600 --> 09:38.720 It's thinking now I will tell you that uh, you'll see, by the way, that the agent has been assigned 09:38.720 --> 09:40.920 the task financial researcher for Tesla. 09:40.960 --> 09:48.140 That shows how that those inputs, those parameters automatically got set as the, the, uh, the agents, 09:48.180 --> 09:48.980 uh, roll. 09:49.300 --> 09:53.020 Now, I'll tell you that the deep sea takes takes its time over this. 09:53.060 --> 09:58.740 It takes a good 30s, I guess, because it's producing quite comprehensive research as part of this. 09:58.740 --> 10:00.620 So we will allow it to do its thing. 10:00.620 --> 10:03.900 Maybe next time we should have them both be grok so that it's nice and fast. 10:04.180 --> 10:05.540 But it's going through. 10:05.660 --> 10:10.740 It's, uh, we've gone off to deep sea guys servers where it is. 10:10.940 --> 10:19.220 The 671 billion parameter model is busy at work, and once it's finished, we're then going to flip 10:19.260 --> 10:26.620 across to our other agent, and our other agent will use grok running in the cloud a 70 billion meta 10:26.620 --> 10:27.180 model. 10:27.340 --> 10:31.340 And then we should get our financial report on Tesla at the end of it. 10:31.340 --> 10:31.980 Here it is. 10:32.020 --> 10:33.980 Grok was so quick we didn't even see it. 10:34.420 --> 10:38.100 There's uh the grok taking the task and completing it. 10:38.140 --> 10:40.100 Here is the result. 10:40.140 --> 10:42.900 Our financial report on Tesla. 10:42.940 --> 10:44.620 It all looks great. 10:44.660 --> 10:46.100 Wasn't that easy?