WEBVTT 00:00.120 --> 00:05.720 All right, so let me now run everything and let me go and run it in debug mode. 00:08.040 --> 00:09.560 So we got a result. 00:09.600 --> 00:11.640 Let's go and check it so we can see. 00:11.680 --> 00:15.280 Now we extracted from structured response. 00:15.280 --> 00:18.360 This is um this is the key here in the results. 00:18.680 --> 00:21.240 And we have here by the way all the messages. 00:21.240 --> 00:25.320 So we have here all the information of of what happened. 00:25.480 --> 00:27.440 So we can really see the flow here. 00:27.440 --> 00:29.040 So first we send the message. 00:29.320 --> 00:34.120 Then the iMessage returned us with a tool call so we can check it out. 00:34.120 --> 00:38.560 We can have here here in the content box we can have here a tool call. 00:38.720 --> 00:42.960 And here interestingly enough we have three tool calls to be made here. 00:43.200 --> 00:51.280 So we invoked the um search for a bunch of times here three times. 00:51.720 --> 00:55.400 Then we go and we have three tool messages. 00:55.400 --> 00:57.240 So we have three tool results. 00:57.480 --> 01:00.600 And we finally go aggregate everything. 01:00.600 --> 01:05.820 And this is what I returned here, which is going to be the structured response. 01:05.820 --> 01:07.220 So we have here the answer. 01:07.620 --> 01:10.900 So this is going to be a string with all the job postings. 01:11.300 --> 01:14.100 Let me show you the sources here. 01:14.860 --> 01:15.220 Right. 01:15.220 --> 01:17.820 So we can see now those are the sources. 01:17.820 --> 01:19.780 Those are the LinkedIn URLs. 01:19.820 --> 01:22.140 Let me go now and continue the run. 01:22.380 --> 01:24.700 And here we simply printed everything. 01:25.180 --> 01:25.540 All right. 01:25.540 --> 01:27.580 So now let me go to transmit here. 01:27.580 --> 01:29.420 And here we have the search agent. 01:30.340 --> 01:33.460 And here we can see now the flow here of the run. 01:33.620 --> 01:37.500 And notice here that here we have lane graph. 01:37.580 --> 01:42.460 So lane graph here is going to be our engine which is going to be running everything. 01:42.460 --> 01:48.500 So it's going to be replacing the agent executor that that we saw here with lane graph primitives. 01:48.660 --> 01:52.420 And we'll be diving very deep into this later in this course. 01:52.820 --> 01:58.620 And because it's going to be using graph now under the hood, it's going to give us the ability to use 01:58.620 --> 02:05.400 the graph debugger, which I'm going to show you in the next video, let's quickly skim the trace. 02:05.600 --> 02:12.080 Now, in contrast to traces that we've seen before, this trace is actually a line graph trace. 02:12.280 --> 02:15.640 So this means it runs in the graph runtime. 02:15.880 --> 02:24.040 Now I'm not going to dive deep into explain any of the graph components here, because we have enough 02:24.040 --> 02:26.040 room to do it later in this course. 02:26.280 --> 02:30.960 I am just wanting to show you what we've already saw in the previous implementation. 02:30.960 --> 02:33.240 I just want to show you the flow here. 02:33.480 --> 02:40.400 And this is a quick introduction and a quick glimpse into the graph trace and the function calling trace 02:40.440 --> 02:44.480 trace which we're going to elaborate on this course in later sections. 02:44.840 --> 02:47.600 So the first call was to open AI. 02:47.960 --> 02:50.280 And notice here we have two tools. 02:50.320 --> 02:52.240 The first tool is the search. 02:52.400 --> 02:56.200 And the second tool is the agent response schema. 02:56.720 --> 03:04.220 So we can see here that the model selected with function calling the tool to be executed to be, to 03:04.220 --> 03:04.940 really search. 03:05.380 --> 03:10.140 And we can see that the output was a bunch of search queries. 03:10.140 --> 03:13.380 So we can see that it wanted to invoke the search. 03:13.540 --> 03:16.380 And it had a bunch of different queries. 03:16.580 --> 03:23.900 The models to tool step here is simply land graph orchestration here, where land Graph now is going 03:23.900 --> 03:32.180 to take every search query and it's going to send it to execute in land graph into different nodes. 03:32.180 --> 03:33.740 Those are primitives in land graphs. 03:33.740 --> 03:36.900 And we're going to be covering that later in this course here. 03:37.180 --> 03:41.740 At this point of the course, I'm not expecting you to know exactly what's happening here. 03:41.740 --> 03:43.940 In each step here of the trace. 03:44.100 --> 03:47.180 You are going to know that later in this course. 03:47.780 --> 03:52.300 Anyways, now we can see that we have here three invocations of the tools. 03:52.540 --> 03:55.820 So those are invocations that are going to run in parallel. 03:56.060 --> 03:59.340 And each one is going to run a different query into villi. 03:59.820 --> 04:06.520 And lastly we have the final call to the model after we have the two results, and the model is going 04:06.520 --> 04:10.960 to decide that it has enough information and it can output us an answer. 04:11.200 --> 04:16.280 And that answer is going to be an agent response object. 04:16.400 --> 04:19.320 So we're going to see that. 04:19.320 --> 04:21.680 Now this is being called as a tool. 04:21.680 --> 04:25.120 And don't worry I will be explaining how everything here is working. 04:25.400 --> 04:32.680 But the tldr of it is that the final response that we're going to be getting is the same agent response 04:32.680 --> 04:39.560 that we got from earlier, but this time it's going to be using function calling under the hood to call 04:39.560 --> 04:46.560 the tool, and also to format us the output as the agent response object. 04:46.720 --> 04:47.160 All right. 04:47.160 --> 04:52.000 So the answer here is going to be here under the structured response here. 04:52.240 --> 04:58.960 And this is going to be an object of agent response that we defined earlier in here popping out. 04:58.960 --> 05:02.080 And I know I'm being pretty vague here in this trace. 05:02.080 --> 05:05.020 And we're not going over it fully. 05:05.020 --> 05:08.460 So we're not going over into every part of this race. 05:08.580 --> 05:15.340 And this is simply because at this point of this course, I think it would be a bit overwhelming introducing 05:15.340 --> 05:17.540 a bunch of landgraaf primitives. 05:17.540 --> 05:23.700 We have plenty of time to learn about landgraaf and diving deep into landgraaf, but before we do that, 05:23.700 --> 05:28.780 we really need to understand how the react algorithm is working underneath the hood. 05:28.780 --> 05:31.940 And this is what we're going to be doing in future sections. 05:31.940 --> 05:38.300 After that, we'll be reviewing function calling, and then it would be the correct time to go and move 05:38.300 --> 05:39.660 into Landgraaf. 05:39.740 --> 05:42.620 So we have a bit of a way to go through here. 05:42.780 --> 05:44.900 And trust me, I know what I'm doing here. 05:44.900 --> 05:52.820 So if you want, you can maybe put a bookmark here on this video and on this trace specifically. 05:53.060 --> 05:58.940 And I promise you that in a couple of sections you will be able to fully read it and understand every 05:58.940 --> 06:00.580 part that's going on here. 06:01.140 --> 06:05.660 So I will be publishing this trace and sharing it in the videos. 06:05.660 --> 06:06.340 Resources.