WEBVTT 00:00.120 --> 00:01.320 Okay, let's do this. 00:01.360 --> 00:02.640 What tool do we think we're going to pick? 00:02.640 --> 00:06.720 We press plus and of course it is going to be exactly the same. 00:06.760 --> 00:07.560 We do super bass. 00:07.560 --> 00:10.600 We want a super bass vector store as our tool. 00:10.640 --> 00:11.480 Up it comes. 00:11.680 --> 00:14.880 Super bass account operation. 00:15.160 --> 00:19.280 Uh, retrieve documents as tool for AI agent. 00:19.280 --> 00:21.160 That sounds like a good one to me. 00:21.360 --> 00:23.680 Uh, let's see the other things. 00:23.800 --> 00:25.560 That's definitely the one, isn't it? 00:25.720 --> 00:27.400 Um, so now a description. 00:27.400 --> 00:32.560 So we need to, to use this to tell the agent when it would want to use use it. 00:32.560 --> 00:46.560 So use this tool to look up any product information that relates to products offered by the company. 00:48.000 --> 00:52.600 Use this tool to look up any information that relates to products offered by the company. 00:52.680 --> 00:56.720 Keep it a bit generic so that this could apply to whatever your company is. 00:56.720 --> 01:00.280 This time I will try not to forget to select a table name that I did last time. 01:00.320 --> 01:02.240 It pops up and we're on knowledge base. 01:02.400 --> 01:07.880 Uh, and uh, um, we'll include the metadata in the, in the result. 01:07.880 --> 01:09.420 There's, there's like a limit thing here. 01:09.420 --> 01:12.100 We might as well allow it to bring back ten documents. 01:12.100 --> 01:13.060 I don't see why not. 01:13.300 --> 01:14.820 Uh, and, uh. 01:14.860 --> 01:15.100 Yeah. 01:15.140 --> 01:15.860 Rerank results. 01:15.860 --> 01:17.420 That's one of the fancy things. 01:17.420 --> 01:25.140 If you want to use an LLM to decide how to reorder what comes back, which which you can look at, uh, 01:25.140 --> 01:26.460 at some of that and experiment with it. 01:26.460 --> 01:27.300 But we're not going to do that. 01:27.300 --> 01:30.060 We're going to keep it keep it vanilla out of the box. 01:30.220 --> 01:32.500 Um, and uh, so there we go. 01:32.780 --> 01:35.340 Uh, here here is our tool. 01:35.340 --> 01:36.100 It looks great. 01:36.100 --> 01:39.940 It's got a red box, presumably because we haven't yet told it what embedding to use. 01:40.100 --> 01:43.260 And you might be thinking, hang on, why do we need to use an embedding? 01:43.260 --> 01:46.660 Haven't we already taken our data and turned it all into vectors? 01:46.660 --> 01:47.780 And we got into the vector store. 01:48.220 --> 01:49.620 But hopefully you know the answer to this. 01:49.660 --> 01:51.340 You're like, ah, come on, we get this. 01:51.340 --> 01:55.220 The reason you need an embedding is because when the user asks a question, we need to vectorize that 01:55.220 --> 01:55.580 question. 01:55.580 --> 01:59.100 Do you remember that like the first line on that on that diagram. 01:59.100 --> 02:04.540 So, uh, yes, it needs to know what embedding model to use, what Vectorizer to use. 02:04.540 --> 02:06.780 And we're using OpenAI embeddings. 02:06.780 --> 02:07.620 There it is. 02:07.860 --> 02:13.380 We are using text embedding three small uh, and uh, there we go. 02:13.420 --> 02:14.380 That's all we need. 02:14.580 --> 02:17.630 And now the angry red thing has gone away. 02:17.670 --> 02:19.670 Everything looks good. 02:19.950 --> 02:23.790 So, you know, presumably there's there's a there's a lot more complex than this. 02:23.790 --> 02:25.070 This is just us getting started. 02:25.110 --> 02:26.110 But no, this is it. 02:26.110 --> 02:26.990 This should be it. 02:27.310 --> 02:27.670 This. 02:27.670 --> 02:29.550 This is what it takes to build a gigantic rag. 02:29.550 --> 02:33.750 And I love the way that this this diagram looks quite similar to the diagram. 02:33.750 --> 02:36.110 I drew the schematic on how it works. 02:36.110 --> 02:41.350 That's the beautiful thing about Nw10 is that you can really sort of draw out there in this canvas, 02:41.470 --> 02:44.910 just, just the way that you would think about it naturally. 02:44.910 --> 02:45.590 Anyway. 02:45.790 --> 02:48.590 It's it's very expressive and simple in that way. 02:49.230 --> 02:49.990 All right. 02:50.150 --> 02:53.070 So I think we should try I think we should try it. 02:53.430 --> 02:54.350 All right, let's do it. 02:54.630 --> 02:54.990 All right. 02:54.990 --> 02:55.510 Here we go. 02:55.510 --> 02:56.070 Hi there. 02:57.910 --> 02:58.790 Everything's off. 02:59.750 --> 03:00.230 Hello. 03:00.230 --> 03:03.590 If you're looking for specific products or have questions about what we have in stock, feel free to 03:03.670 --> 03:03.950 ask. 03:03.990 --> 03:04.270 All right. 03:04.270 --> 03:07.870 Because it read the tool description so it knows what we're dealing with. 03:07.870 --> 03:09.230 Everything seems to work okay. 03:09.270 --> 03:15.270 So let me say uh, do you, uh, sell any keyboards. 03:17.190 --> 03:17.950 Let's see what happens. 03:17.950 --> 03:20.590 You can see it's doing a query, a lookup a lookup. 03:20.630 --> 03:21.670 Stuff's coming back. 03:22.590 --> 03:23.750 It's going to answer us. 03:23.870 --> 03:24.270 Haha. 03:24.510 --> 03:28.610 Yes, yes, we have a wide selection of keyboards to suit different needs. 03:28.610 --> 03:30.850 Here are some of our popular options. 03:30.970 --> 03:34.490 Gaming and Performance Pro RGB gaming keyboard. 03:34.530 --> 03:36.330 And here they all are. 03:36.730 --> 03:37.130 Wow. 03:37.130 --> 03:37.810 There's a lot of them. 03:37.810 --> 03:39.210 And see that it tells us the price. 03:39.210 --> 03:42.530 I was going to ask it for the price, but it's got the price and I do. 03:42.570 --> 03:45.010 Luckily I was gonna go back and check at the database. 03:45.010 --> 03:47.930 It's really there, but I actually recognize a couple of these. 03:47.930 --> 03:53.570 So it's nice and it's given it a description, but it's grouped it and organized it its way. 03:53.570 --> 03:58.690 And it's clear that because we've got that many, uh, coming back that, that, uh, it's able to give 03:58.690 --> 04:02.610 us the full range of keyboards, which is terrific. 04:02.610 --> 04:14.570 So, um, uh, how much does it cost to purchase more memory for my computer? 04:15.890 --> 04:17.450 Let's see what it says. 04:18.330 --> 04:19.490 It's looking it up. 04:19.650 --> 04:20.810 It's looking it up. 04:23.570 --> 04:24.290 And here we go. 04:24.330 --> 04:28.570 The cost from the desktop or laptop, whether your system is blah, blah, blah, blah, blah. 04:28.970 --> 04:30.210 And here it is. 04:30.330 --> 04:33.940 And this is all the results that have come from the knowledge base. 04:34.060 --> 04:36.460 And so here you have it. 04:36.460 --> 04:38.580 I don't know if you thought this part was going to be complicated. 04:38.580 --> 04:40.420 It's really not this. 04:40.460 --> 04:45.580 I feel like this is when the simplicity of nw10 really comes to the forefront. 04:45.740 --> 04:49.020 Building this rag pipeline was so easy. 04:49.020 --> 04:54.580 And again, remember the big aha moment is when you realize how scalable this is. 04:54.580 --> 04:58.580 It wouldn't be such a big deal if I told you that I could build a question answerer that could answer 04:58.580 --> 05:02.980 questions about 60 products we've got, because you could just paste that into ChatGPT. 05:03.220 --> 05:05.220 The point is, it doesn't need to be 60. 05:05.260 --> 05:08.620 It could be 60,000, it could be 600,000. 05:08.660 --> 05:09.380 This would work. 05:09.380 --> 05:11.140 It would ingest, it would vectorize. 05:11.140 --> 05:16.020 It would pick the nearest neighbors, and it would use this to be able to answer questions as if it 05:16.020 --> 05:17.100 knew the whole lot. 05:17.140 --> 05:24.380 And that that is the the true power of Rag and the power of Nan is that it made it all so easy. 05:24.380 --> 05:25.540 But you know the deal. 05:25.540 --> 05:30.180 You know that we're not here to to chat in this chat box right here. 05:30.180 --> 05:32.420 That's not the point of week two, is it? 05:32.660 --> 05:34.020 Not at all. 05:34.060 --> 05:35.220 Let's hide the chat. 05:35.260 --> 05:37.620 Let's come here and delete that. 05:37.820 --> 05:40.340 We don't have any interest whatsoever. 05:40.340 --> 05:42.580 It's time for us to build our voice agent.