WEBVTT 00:00.160 --> 00:02.360 Hello everyone and welcome. 00:02.800 --> 00:08.960 In this video we will continue where we left off in our previous video where we created an multimodal 00:09.000 --> 00:11.720 Agentic React architecture pipeline. 00:12.200 --> 00:15.080 In this video we will continue from there. 00:15.320 --> 00:21.240 We will create another tool which will get us details about the seven wonders of the world, and we 00:21.240 --> 00:22.640 will ask queries. 00:22.840 --> 00:28.520 Depending on the question, the agent will pick up the particular tool and answer questions. 00:28.880 --> 00:31.440 So now I have to first index the data. 00:31.560 --> 00:38.600 We will start by getting the data from the Hugging face seven wonder space that they have created, 00:38.800 --> 00:41.240 and then index them in our database. 00:41.520 --> 00:45.680 So now this is the data sets that I'm loading from the hugging face. 00:45.800 --> 00:49.560 I'm printing the data here to learn how the data looks like. 00:50.040 --> 00:53.920 So in this case here this is a document from Hugging Face. 00:54.240 --> 00:59.160 It has a document ID and the content about the different wonders of the world. 00:59.640 --> 01:04.360 Now I'll go ahead and index a document in memory document store. 01:04.980 --> 01:08.260 So here we created an in-memory document store. 01:08.740 --> 01:12.540 For that I will have to use the sentence transformer. 01:12.820 --> 01:16.700 All mini LM model will download the model. 01:17.020 --> 01:19.620 We'll embed the document with this embedder. 01:19.980 --> 01:22.580 And then we'll write the document to the store. 01:22.620 --> 01:29.820 Once it's embedded and then once the documents are embedded in the document store, we initialize it 01:29.820 --> 01:32.380 with the in-memory retriever object. 01:32.620 --> 01:37.020 You can either do this or you can create a pipeline out of this. 01:37.540 --> 01:39.780 Whatever works best for you. 01:40.460 --> 01:45.980 Now that we have the retriever ready, we need to build a retriever pipeline for our questions. 01:46.340 --> 01:54.540 So for that we need a sentence transformer, text retriever, in-memory embedding retriever, and a 01:54.540 --> 01:55.620 prompt builder. 01:56.260 --> 02:05.580 So I have three different prompt components initialized in memory embedding retriever sentence transformer 02:06.140 --> 02:09.050 text Embedder to embed questions. 02:09.850 --> 02:16.170 This will help us fetch the top K document, and then prompt Builder would help construct the prompt 02:16.170 --> 02:18.250 that our agent would eventually use. 02:18.530 --> 02:27.490 So I created this pipeline seven Wonders Pipeline and added components text Embedder retriever and Prompt 02:27.490 --> 02:28.970 Builder to the pipeline. 02:29.290 --> 02:36.450 And then I'm connecting the pipeline as text Embedder to Retriever retriever to the prompt builder. 02:37.170 --> 02:41.490 Now let's go to the next step here which is running the retrieval pipeline. 02:41.970 --> 02:47.970 I'm running the Seven Wonders pipeline with text Embedder as the embedding component. 02:48.490 --> 02:53.770 The question that I'm asking here is what does road Statue look like? 02:54.010 --> 02:56.330 And then I'm printing the response. 02:56.810 --> 03:00.730 Let me go ahead and run the code blocks that I might have missed in the video. 03:01.210 --> 03:02.570 So we went through this. 03:02.930 --> 03:11.450 There are a 151 documents that it wrote creating and initializing the components, adding it to the 03:11.450 --> 03:12.330 pipeline. 03:12.830 --> 03:14.710 and then running the pipeline. 03:15.230 --> 03:18.470 So I did get the response here about the road statue. 03:18.870 --> 03:20.990 Let's confirm how this looks like. 03:21.430 --> 03:23.470 So here's the content that I got. 03:23.830 --> 03:27.790 And if you notice here it is relevant to the road statue. 03:28.310 --> 03:33.590 The next step is to create the tool with Seven Wonders pipeline. 03:34.110 --> 03:43.030 So for that I have imported the Fast Drag Tools library with document with Image Query tool as a the 03:43.030 --> 03:44.590 tool that I'll use here. 03:44.670 --> 03:50.830 I'm giving the name as seven Wonders tool, and then the description is useful for when you need to 03:50.870 --> 03:53.270 retrieve Fact of seven wonders. 03:53.790 --> 03:57.710 It can give you information about one wonder per query. 03:58.350 --> 04:00.070 Pass the wonder as input. 04:00.070 --> 04:06.550 This is a very important step where it uses the name and the description of its functionality to decide 04:06.550 --> 04:08.070 which tool to invoke. 04:08.550 --> 04:11.510 Agent uses this information appropriately. 04:11.910 --> 04:17.980 So please make sure that the description and the name is very relevant to the task. 04:18.340 --> 04:22.900 And then this is the pipeline that we have specified up here. 04:23.300 --> 04:23.820 Right. 04:24.300 --> 04:26.780 Let's execute this okay. 04:27.180 --> 04:29.620 So we got the seven wonder as a tool. 04:29.620 --> 04:30.180 Ready. 04:30.780 --> 04:32.660 Now let's go ahead and invoke this. 04:32.940 --> 04:38.820 So here is the next piece of code that says seven wonder tool that we created up here. 04:39.260 --> 04:41.260 Road statue is the input. 04:41.580 --> 04:45.660 Then I get the tools result and then I'll print the response. 04:45.900 --> 04:47.860 So we got the response. 04:48.100 --> 04:54.300 And if you notice here we have the information about road statue and the details that we got from the 04:54.300 --> 04:55.140 pipeline. 04:55.420 --> 04:58.420 I went ahead and tried a different option for the pipeline. 04:58.700 --> 05:03.940 In this case I wrote pyramid and it came back with the information of pyramid. 05:04.380 --> 05:09.660 Now let's go ahead and use the generator to come up with the entire agent experience. 05:10.100 --> 05:15.100 So here I have imported the agent and tools manager from the FastTrack. 05:15.580 --> 05:21.480 Along with that, I am also importing the conversation Memory with the agent class. 05:21.840 --> 05:24.040 Here, there's a generator. 05:24.280 --> 05:26.240 There's a prompt template tool. 05:26.560 --> 05:29.960 It is different in our case, which is seven wonders tool. 05:29.960 --> 05:30.840 Conversation. 05:30.840 --> 05:31.480 Memory. 05:31.760 --> 05:34.560 We pass it to the generator and we run this. 05:35.080 --> 05:37.320 How does road statue look like? 05:37.600 --> 05:39.120 Let's execute this. 05:39.760 --> 05:44.480 So the thought is I need to find out the appearance of the road statue. 05:44.800 --> 05:47.720 It's using the tool seven wonder tool. 05:48.040 --> 05:50.160 The input is road statue. 05:50.160 --> 05:53.920 And then it gave us the response back from the pipeline. 05:54.120 --> 05:58.720 And this is generating the response from the content. 05:59.280 --> 06:02.360 Now what if we run two tools at the same time. 06:03.440 --> 06:10.840 So in this case what we need to do is we need to add one more tool that we created in the previous video. 06:11.200 --> 06:15.120 The tool was nutrition tool that we created in the previous video. 06:15.520 --> 06:19.320 Let's go ahead and ask a question about how does Road Statue look like? 06:19.560 --> 06:23.240 And the agent has two tools to evaluate and pick one. 06:23.550 --> 06:27.310 Let's see if the agent can do justice to the tool's invocation. 06:27.710 --> 06:32.550 So in this case, the user is asking for the appearance of the road statue. 06:32.910 --> 06:40.070 Then it picked up seven Wonders tool, and we got all the information from the retrieval pipeline to 06:40.110 --> 06:43.950 whatever we had before, and then it summarized it. 06:44.270 --> 06:46.230 I will use another query. 06:46.550 --> 06:49.790 How much protein does protein bar have? 06:50.270 --> 06:53.310 And then execute this one more time. 06:53.790 --> 06:57.790 So the thought is the protein content in the protein bar. 06:58.230 --> 07:04.310 It invoked the nutrition tool and it invoked the necessary prompt with image. 07:04.870 --> 07:07.630 It got the information from the content. 07:08.030 --> 07:12.990 And then the final answer was that protein bar has 14g of protein. 07:13.190 --> 07:14.150 That's amazing. 07:14.630 --> 07:19.030 It was able to invoke the right tool to the right question that was asked. 07:19.430 --> 07:24.870 I hope you learned Multiagent application with Multi-hop using React architecture. 07:25.110 --> 07:28.070 Thank you and I'll see you in the next video.