WEBVTT 00:00:00.000 --> 00:00:03.140 The surprising reality of modern AI conversations 00:00:03.140 --> 00:00:06.360 is that they aren't just chats anymore. They're 00:00:06.360 --> 00:00:10.619 becoming transactions. Imagine an AI agent checking 00:00:10.619 --> 00:00:12.679 your actual calendar, maybe booking a 30 -minute 00:00:12.679 --> 00:00:14.839 consultation and doing it all with a really professional 00:00:14.839 --> 00:00:17.440 voice in real time, seamlessly. This isn't some 00:00:17.440 --> 00:00:19.359 future tech concept. It's happening right now. 00:00:19.480 --> 00:00:21.420 And this deep dive, well, it unpacks how you 00:00:21.420 --> 00:00:23.379 can build this exact system yourself. Exactly. 00:00:24.000 --> 00:00:25.800 Welcome, everyone, to the Deep Dive. This is 00:00:25.800 --> 00:00:27.820 where we tear into the technical guides so you 00:00:27.820 --> 00:00:30.160 don't have to. And you shared a really comprehensive 00:00:30.160 --> 00:00:33.820 blueprint on integrating N8N and 11LAPS, basically 00:00:33.820 --> 00:00:35.979 charting out the architecture for a powerful 00:00:35.979 --> 00:00:39.100 conversational agent. So our mission today is 00:00:39.100 --> 00:00:41.579 to shortcut that build process for you. We're 00:00:41.579 --> 00:00:43.340 going to extract the critical architecture, the 00:00:43.340 --> 00:00:45.539 configuration steps you actually need for a fully 00:00:45.539 --> 00:00:47.560 functional voice assistant. We'll kick things 00:00:47.560 --> 00:00:49.420 off by looking at that conversational example, 00:00:49.600 --> 00:00:51.579 which is pretty powerful. Then we'll dissect 00:00:51.579 --> 00:00:53.259 the architecture, you know, the brains and the 00:00:53.259 --> 00:00:55.840 hands of the operation. And finally, we'll explore 00:00:55.840 --> 00:00:58.659 customization, some advanced integrations, and 00:00:58.659 --> 00:01:00.340 importantly, how you can turn this knowledge 00:01:00.340 --> 00:01:03.159 into a serious business advantage. Okay, let's 00:01:03.159 --> 00:01:06.409 unpack this then. The core idea here. It seems 00:01:06.409 --> 00:01:09.030 to be combining powerful NEN workflow automation 00:01:09.030 --> 00:01:12.230 with the, well, the advanced realism of Eleven 00:01:12.230 --> 00:01:14.549 Labs voice tech. When you put those two together, 00:01:14.650 --> 00:01:16.689 you create what they're calling a no code voice 00:01:16.689 --> 00:01:19.590 AI agent. But we should probably be real here. 00:01:19.689 --> 00:01:21.750 When they say no code, what they really mean 00:01:21.750 --> 00:01:24.250 is the complex API stuff, the orchestration. 00:01:24.430 --> 00:01:26.969 It's abstracted away into visual nodes. You still 00:01:26.969 --> 00:01:28.890 need to understand the logic. That's a critical 00:01:28.890 --> 00:01:30.950 distinction. Yeah. It's more like low code power, 00:01:31.049 --> 00:01:33.810 not exactly zero code magic. Yeah. But the result. 00:01:34.170 --> 00:01:36.230 It's transformative. I mean, this thing can operate 00:01:36.230 --> 00:01:39.989 24 -7, handling pretty complex customer inquiries 00:01:39.989 --> 00:01:42.310 without needing a human to step in. What's really 00:01:42.310 --> 00:01:44.709 fascinating, I think, is watching that intelligence 00:01:44.709 --> 00:01:46.730 actually work. Let's maybe walk through the Jarvis 00:01:46.730 --> 00:01:49.569 demo scenario they mentioned. So a customer asked, 00:01:49.689 --> 00:01:51.969 can you book a consultation call tomorrow? It 00:01:51.969 --> 00:01:53.530 sounds like a simple natural language request. 00:01:53.810 --> 00:01:56.150 But the key thing is the AI agent immediately 00:01:56.150 --> 00:01:58.909 recognizes the intent behind it, the intent to 00:01:58.909 --> 00:02:00.969 perform an action. Right. It doesn't just guess. 00:02:01.109 --> 00:02:03.700 This is where the magic of function. calling 00:02:03.700 --> 00:02:06.180 comes into play. The underlying large language 00:02:06.180 --> 00:02:08.560 model, the LLM, it's configured with definitions 00:02:08.560 --> 00:02:11.900 of available tools like get events. So it realizes, 00:02:11.979 --> 00:02:13.860 okay, the customer needs me to run this tool 00:02:13.860 --> 00:02:16.259 before I can answer properly. And it hits the 00:02:16.259 --> 00:02:19.699 live calendar, gets the specifics. And then it 00:02:19.699 --> 00:02:21.620 speaks naturally saying something like, tomorrow 00:02:21.620 --> 00:02:24.960 AI fire is busy from 9 -0 -0 -0 -0 -0 in the 00:02:24.960 --> 00:02:29.669 morning, but open. Between 10 .0000. The conversational 00:02:29.669 --> 00:02:32.090 flow is really key there. Once the time slot 00:02:32.090 --> 00:02:34.629 is picked, the AI just automatically collects 00:02:34.629 --> 00:02:37.530 the necessary info, you know, name, email, phone 00:02:37.530 --> 00:02:39.430 number, and confirms all the booking details. 00:02:39.770 --> 00:02:42.050 And it goes way beyond just booking. This setup 00:02:42.050 --> 00:02:43.969 shows off what they call an intelligent knowledge 00:02:43.969 --> 00:02:46.590 base. So if the customer switches gears and asks 00:02:46.590 --> 00:02:49.069 about pricing, the agent can instantly pull up 00:02:49.069 --> 00:02:51.310 the accurate information. Okay, for an initial 00:02:51.310 --> 00:02:54.150 30 -minute consultation, the flat fee is $150. 00:02:55.439 --> 00:02:57.460 It's context switching and real -time execution 00:02:57.460 --> 00:03:00.259 happening together. It's actively checking and 00:03:00.259 --> 00:03:02.759 booking right into a Google Calendar, all in, 00:03:02.780 --> 00:03:04.900 like a fraction of a second. Okay, let's zoom 00:03:04.900 --> 00:03:07.620 out for a second. How does the AI actually know 00:03:07.620 --> 00:03:09.919 it needs to run some external code just because 00:03:09.919 --> 00:03:13.240 the user asked to book a meeting? Well, the system 00:03:13.240 --> 00:03:16.240 prompt explicitly defines the actions the LLM 00:03:16.240 --> 00:03:18.159 is allowed to take. It's given specific instructions. 00:03:18.560 --> 00:03:21.340 Ah, okay. So the prompt itself tells it which 00:03:21.340 --> 00:03:23.439 tools are available for which requests. Exactly. 00:03:23.500 --> 00:03:25.939 Now let's dissect the components. The whole system, 00:03:26.000 --> 00:03:28.639 it's a remarkably efficient machine, really built 00:03:28.639 --> 00:03:31.280 on three main layers. First, you've got the front 00:03:31.280 --> 00:03:33.759 -end voice interface. That's 11 labs. It's the 00:03:33.759 --> 00:03:36.879 voice. It's the ears. It handles the speech to 00:03:36.879 --> 00:03:39.599 text, generates that realistic voice output, 00:03:39.780 --> 00:03:42.360 and crucially, it recognizes the action intent. 00:03:42.659 --> 00:03:45.300 Okay. Then you have the backend workflow engine, 00:03:45.439 --> 00:03:47.900 and that's N8n. This is kind of the operational 00:03:47.900 --> 00:03:50.719 brain and the hands, too. It manages all the 00:03:50.719 --> 00:03:52.939 business logic, connects securely to your different 00:03:52.939 --> 00:03:56.000 apps, processes data, and controls those real 00:03:56.000 --> 00:03:59.150 -world actions using APIs. Right. And the sort 00:03:59.150 --> 00:04:01.069 of invisible glue holding them together. That's 00:04:01.069 --> 00:04:03.250 the integration layer. And it relies entirely 00:04:03.250 --> 00:04:05.530 on webhooks. You can think of a webhook like 00:04:05.530 --> 00:04:08.750 a secure automated message, like a text message 00:04:08.750 --> 00:04:11.349 sent from one application to another when a specific 00:04:11.349 --> 00:04:13.729 event happens. In this case, it happens when 00:04:13.729 --> 00:04:16.250 the 11 Labs agent decides, OK, I need to use 00:04:16.250 --> 00:04:18.449 one of my tools. All right. Let's trace that 00:04:18.449 --> 00:04:20.430 lightning fast journey of a customer command. 00:04:20.610 --> 00:04:22.990 We can imagine it like stacking Lego blocks of 00:04:22.990 --> 00:04:26.389 data or maybe a relay race. So step one, the 00:04:26.389 --> 00:04:29.879 customer speaks. Voice input. 11 Labs converts 00:04:29.879 --> 00:04:32.060 that speech to text, figures out the intent, 00:04:32.339 --> 00:04:34.879 and this seems like the critical part triggers 00:04:34.879 --> 00:04:37.980 a specific tool activation. That tool activation, 00:04:38.160 --> 00:04:40.600 it's basically a package of data, payload, sent 00:04:40.600 --> 00:04:43.259 via that webhook. That payload hits the N8 Play 00:04:43.259 --> 00:04:46.420 ON server, which kicks off step four. The N8 00:04:46.420 --> 00:04:48.439 PES workflow executes. It runs its automation, 00:04:48.660 --> 00:04:50.720 maybe connects to Google Calendar, checks the 00:04:50.720 --> 00:04:53.800 availability. Step five, N8N puts together the 00:04:53.800 --> 00:04:55.920 answer and sends the text response back to 11 00:04:55.920 --> 00:04:59.730 Labs. And then finally, step six, 11 Labs converts 00:04:59.730 --> 00:05:02.310 that text back into natural sounding speech for 00:05:02.310 --> 00:05:05.490 the customer. The speed is actually quite shocking 00:05:05.490 --> 00:05:07.829 when you hear it. It feels like just one seamless, 00:05:07.850 --> 00:05:10.910 immediate conversation. Yeah, it's fast. So if 00:05:10.910 --> 00:05:13.110 we look at that connection point, that web hook. 00:05:13.629 --> 00:05:16.610 What structural piece really makes this whole 00:05:16.610 --> 00:05:19.329 data relay system stable and secure enough for 00:05:19.329 --> 00:05:21.569 live transactions? Well, based on the description, 00:05:21.790 --> 00:05:24.329 the webhook trigger acts as that digital bridge. 00:05:24.529 --> 00:05:28.209 It's the secure and required API endpoint. Okay, 00:05:28.310 --> 00:05:30.589 that makes sense. Now, let's talk about building 00:05:30.589 --> 00:05:33.129 the actual intelligence, the foundation, which 00:05:33.129 --> 00:05:36.810 lives within AAN. Most professional builds, they 00:05:36.810 --> 00:05:38.769 don't start... totally from scratch, right? They 00:05:38.769 --> 00:05:41.550 often use a pre -built JSON blueprint, like a 00:05:41.550 --> 00:05:43.329 starting template for the workflow structure. 00:05:43.589 --> 00:05:46.089 This definitely saves time, but it also brings 00:05:46.089 --> 00:05:47.870 up some important questions, I think. It does. 00:05:48.009 --> 00:05:49.790 Like if I'm using a pre -built blueprint someone 00:05:49.790 --> 00:05:52.990 else made, am I just copying their security configuration 00:05:52.990 --> 00:05:56.029 without thinking? How do I actually ensure data 00:05:56.029 --> 00:05:58.629 safety when I'm linking my live Google Calendar, 00:05:58.790 --> 00:06:01.269 putting my API key in there? That feels like 00:06:01.269 --> 00:06:03.250 a real concern in this sort of low -code world. 00:06:03.680 --> 00:06:05.720 Absolutely. Yeah, you really have to scrutinize 00:06:05.720 --> 00:06:08.300 those critical nodes. Within that NEI Gen workflow 00:06:08.300 --> 00:06:11.500 structure, you basically see two key players. 00:06:11.720 --> 00:06:14.500 First is the webhook trigger node. That's just 00:06:14.500 --> 00:06:16.779 the dedicated entry point. It's the ears of the 00:06:16.779 --> 00:06:18.740 workflow just sitting there waiting for that 00:06:18.740 --> 00:06:21.379 signal from 11 labs. But the real intelligence, 00:06:21.540 --> 00:06:24.079 the brainpower, that's in the AI agent node. 00:06:24.459 --> 00:06:26.879 That's the complex decision maker. It takes that 00:06:26.879 --> 00:06:29.600 text payload from a webhook, processes the language 00:06:29.600 --> 00:06:32.079 using the LLM, and then decides, okay, does this 00:06:32.079 --> 00:06:34.600 request mean I need to connect to the Google 00:06:34.600 --> 00:06:37.279 Calendar node, or can I just give a static answer 00:06:37.279 --> 00:06:39.970 like the pricing info? And probably the most 00:06:39.970 --> 00:06:42.329 crucial step in configuring this whole setup 00:06:42.329 --> 00:06:44.889 is actually writing the AI's job description. 00:06:45.170 --> 00:06:47.750 That's the system prompt configuration. This 00:06:47.750 --> 00:06:49.930 defines the agent's personality. You know, is 00:06:49.930 --> 00:06:52.470 it helpful, friendly, professional? And it sets 00:06:52.470 --> 00:06:55.230 the rules for how it uses its tools. Right. And 00:06:55.230 --> 00:06:57.050 this is where it gets nuanced, isn't it? You 00:06:57.050 --> 00:06:59.310 have to give really explicit instructions like, 00:06:59.389 --> 00:07:01.670 if an appointment is being requested, you must 00:07:01.670 --> 00:07:04.149 first check the get events tool and report the 00:07:04.149 --> 00:07:06.670 availability before you suggest booking. That 00:07:06.670 --> 00:07:09.339 kind of strict guidance seems vital. You know, 00:07:09.360 --> 00:07:11.199 I still wrestle with prompt drift myself sometimes. 00:07:11.839 --> 00:07:14.199 Ensuring that initial personality and those strict 00:07:14.199 --> 00:07:16.720 rules stay consistent when you get into really 00:07:16.720 --> 00:07:19.220 complex multi -turn conversations. It's tricky. 00:07:19.560 --> 00:07:21.819 oh it's a huge problem especially under what 00:07:21.819 --> 00:07:24.540 you might call conversational stress to help 00:07:24.540 --> 00:07:26.920 mitigate that prompt drift the source material 00:07:26.920 --> 00:07:30.040 really emphasizes using specific structural cues 00:07:30.040 --> 00:07:34.740 things like xml tags or forcing json output parsing 00:07:34.740 --> 00:07:37.459 basically you're telling the llm exactly how 00:07:37.459 --> 00:07:40.199 to process the information and respond not just 00:07:40.199 --> 00:07:43.699 what to process more structure okay then we shift 00:07:43.699 --> 00:07:46.060 over to the 11 lab side And that's where we define 00:07:46.060 --> 00:07:48.699 the agent's actual voice persona. You give the 00:07:48.699 --> 00:07:50.740 agent a memorable name, let's say business assistant, 00:07:50.920 --> 00:07:52.819 and you develop the system prompt there too. 00:07:53.199 --> 00:07:55.300 But this prompt seems more focused on the voice 00:07:55.300 --> 00:07:57.699 delivery, the tone, and setting guardrails like, 00:07:57.800 --> 00:08:00.160 do not share any personal or financial details 00:08:00.160 --> 00:08:02.180 except what's absolutely needed for booking. 00:08:02.569 --> 00:08:04.189 Exactly. And this is where the action really 00:08:04.189 --> 00:08:06.930 gets connected. You take that webhook URL you 00:08:06.930 --> 00:08:09.649 got from your N8n setup and you link it directly 00:08:09.649 --> 00:08:12.290 to the tool definitions, get available slots 00:08:12.290 --> 00:08:15.449 and book meeting right inside 11 Labs. That link, 00:08:15.529 --> 00:08:18.149 that specific connection is the final essential 00:08:18.149 --> 00:08:20.870 piece for triggering real action directly from 00:08:20.870 --> 00:08:23.170 a customer's voice command. So if we've set this 00:08:23.170 --> 00:08:25.569 all up correctly, what's the single command or 00:08:25.569 --> 00:08:28.050 instruction within that system prompt that ensures 00:08:28.050 --> 00:08:31.170 the agent acts responsibly and doesn't say double 00:08:31.170 --> 00:08:33.870 book a time slot? the system prompt mandates 00:08:33.870 --> 00:08:36.350 using the get available slots tool before it's 00:08:36.350 --> 00:08:38.669 allowed to invoke the book meeting tool sequence 00:08:38.669 --> 00:08:41.929 matters mid -roll sponsor read placeholder back 00:08:41.929 --> 00:08:45.230 all right phase three this is all about fine 00:08:45.230 --> 00:08:47.980 tuning Making it resilient, ready for scale, 00:08:48.080 --> 00:08:50.039 you absolutely have to address the possibility 00:08:50.039 --> 00:08:53.159 of webhook timeout issues. What happens if N8n 00:08:53.159 --> 00:08:55.460 takes too long to respond? The source suggests 00:08:55.460 --> 00:08:57.279 implementing things like please wait a moment, 00:08:57.379 --> 00:08:59.440 messages in the agent's responses during those 00:08:59.440 --> 00:09:02.200 processing delays, and also setting up retry 00:09:02.200 --> 00:09:05.659 systems within the N8n workflow itself. And then 00:09:05.659 --> 00:09:08.580 you optimize the voice quality itself. Selecting 00:09:08.580 --> 00:09:10.519 a professional sounding model, maybe tweaking 00:09:10.519 --> 00:09:12.419 the speech rate, making sure the voice is clear, 00:09:12.480 --> 00:09:14.559 understandable, even if the user has a heavy 00:09:14.559 --> 00:09:17.580 accent. That attention to detail, it seems like 00:09:17.580 --> 00:09:19.620 it would dramatically increase user trust, wouldn't 00:09:19.620 --> 00:09:21.940 it? Definitely. Now, here's where it gets really 00:09:21.940 --> 00:09:24.620 interesting, I think. Scaling up requires moving 00:09:24.620 --> 00:09:28.039 beyond just basic knowledge -based lookups. And 00:09:28.039 --> 00:09:30.659 that naturally leads us to vector database integration. 00:09:31.100 --> 00:09:32.919 I'm curious about that. What core limitation? 00:09:33.639 --> 00:09:36.539 does a vector database actually solve that a 00:09:36.539 --> 00:09:39.740 standard say relational database just can't handle 00:09:39.740 --> 00:09:41.720 in this context well it fundamentally solves 00:09:41.720 --> 00:09:44.360 the problem of meaning versus just keywords think 00:09:44.360 --> 00:09:46.299 about it if a customer asks something complex 00:09:46.299 --> 00:09:48.759 like what happens if i miss a payment and need 00:09:48.759 --> 00:09:51.700 to adjust my schedule A simple keyword search 00:09:51.700 --> 00:09:53.779 might completely fail. That doesn't capture the 00:09:53.779 --> 00:09:57.200 nuance. But a vector database, maybe using tools 00:09:57.200 --> 00:10:00.340 like Pinecone, allows the NAN workflow to search 00:10:00.340 --> 00:10:02.580 by semantic meaning, the underlying concept, 00:10:02.820 --> 00:10:05.639 providing much deeper context for highly personalized 00:10:05.639 --> 00:10:08.620 customer experiences. Whoa, okay, imagine scaling 00:10:08.620 --> 00:10:11.019 that kind of system. With a vector database, 00:10:11.320 --> 00:10:14.120 handling maybe a billion complex personalized 00:10:14.120 --> 00:10:17.039 queries simultaneously, constantly recalling 00:10:17.039 --> 00:10:19.740 past interactions, relevant context, all without 00:10:19.740 --> 00:10:22.600 lag, it's pretty profound. And that depth is 00:10:22.600 --> 00:10:24.720 what unleashes huge business value through what 00:10:24.720 --> 00:10:27.100 they call multi -tool expansion. Once you have 00:10:27.100 --> 00:10:30.259 NANN running these workflows, the agent can integrate 00:10:30.259 --> 00:10:33.919 with almost anything with an API. CRMs like Salesforce 00:10:33.919 --> 00:10:36.820 or HubSpot handle payment processing via Stripe, 00:10:36.820 --> 00:10:40.200 send out SMS notifications. Think about the applications. 00:10:40.899 --> 00:10:42.919 Revolutionizing law firms by having the voice 00:10:42.919 --> 00:10:45.360 agent collect structured initial case information 00:10:45.360 --> 00:10:48.320 or medical practices handling initial insurance 00:10:48.320 --> 00:10:50.860 verification automatically. This really explains 00:10:50.860 --> 00:10:52.759 why it's not just an internal efficiency tool. 00:10:52.860 --> 00:10:55.379 It becomes a potential revenue stream. Businesses 00:10:55.379 --> 00:10:57.620 could actually offer voice AI agent development 00:10:57.620 --> 00:11:00.340 as a service. You charge for basic setups, maybe 00:11:00.340 --> 00:11:03.159 $500 to $1 ,500 one time, or perhaps ongoing 00:11:03.159 --> 00:11:05.480 monthly maintenance fees. Yeah. And to make that 00:11:05.480 --> 00:11:07.700 profitable, especially at scale, you absolutely 00:11:07.700 --> 00:11:10.990 need performance optimization. means using techniques 00:11:10.990 --> 00:11:13.289 like load balancing, spreading high volumes of 00:11:13.289 --> 00:11:16.289 traffic across multiple N8N servers so no single 00:11:16.289 --> 00:11:19.649 one gets overloaded, and caching, which is basically 00:11:19.649 --> 00:11:21.830 saving the results of frequent lookups, like 00:11:21.830 --> 00:11:24.389 checking calendar availability, to reduce those 00:11:24.389 --> 00:11:26.950 expensive, time -consuming API calls back to 00:11:26.950 --> 00:11:29.169 Google Calendar or other services. MARK MIRCHANDANI, 00:11:29.169 --> 00:11:30.870 Okay, stepping back again, if someone builds 00:11:30.870 --> 00:11:33.809 this capability today, what's the single biggest 00:11:33.809 --> 00:11:36.490 competitive advantage this level of deep conversational 00:11:36.490 --> 00:11:38.549 integration provides them? I think it offers 00:11:38.549 --> 00:11:41.330 a fully professional, 24 -7 conversational customer 00:11:41.330 --> 00:11:44.230 experience that really sets you apart. It positions 00:11:44.230 --> 00:11:46.450 you as a first mover in leveraging this tech 00:11:46.450 --> 00:11:49.610 effectively. So wrapping this up, what's it all 00:11:49.610 --> 00:11:52.809 really mean? It feels like the future of business 00:11:52.809 --> 00:11:54.730 interaction is increasingly built on these structured 00:11:54.730 --> 00:11:57.570 conversations. And that conversation relies on 00:11:57.570 --> 00:12:00.809 two critical pillars we discussed. The sophisticated 00:12:00.809 --> 00:12:03.710 voice interface, like 11 Labs, and the structured 00:12:03.710 --> 00:12:06.200 logic engine, like NEN. The crucial takeaway 00:12:06.200 --> 00:12:08.240 for me is that combination of, let's call it 00:12:08.240 --> 00:12:09.980 no -code accessibility, at least in terms of 00:12:09.980 --> 00:12:12.759 building the visual workflow, with truly professional 00:12:12.759 --> 00:12:15.919 real -time integration into core business systems 00:12:15.919 --> 00:12:19.360 like calendars, CRMs, databases. Yeah, and if 00:12:19.360 --> 00:12:21.620 we connect this to the bigger picture, as these 00:12:21.620 --> 00:12:24.360 AI agents become genuinely indistinguishable 00:12:24.360 --> 00:12:27.279 from humans in conversation, the next frontier 00:12:27.279 --> 00:12:29.659 isn't just what the agent says or what actions 00:12:29.659 --> 00:12:31.500 it can take. It's going to be how it adapts its 00:12:31.500 --> 00:12:34.679 tone. Imagine an agent using advanced 11 - Labs 00:12:34.679 --> 00:12:37.120 features to actually detect frustration in a 00:12:37.120 --> 00:12:39.720 customer's voice and instantly triggering a specific 00:12:39.720 --> 00:12:42.360 N8N workflow tool that changes its conversational 00:12:42.360 --> 00:12:44.879 pacing, maybe lowers its volume, shifts to a 00:12:44.879 --> 00:12:47.039 more empathetic service style. That's the next 00:12:47.039 --> 00:12:50.179 level. That really does move the AI from just 00:12:50.179 --> 00:12:52.580 being a transactional tool towards becoming more 00:12:52.580 --> 00:12:55.460 of a relationship manager. We definitely encourage 00:12:55.460 --> 00:12:59.259 you, the listener, to explore this concept, building 00:12:59.259 --> 00:13:02.559 these highly specific tool -integrated AI personas 00:13:02.559 --> 00:13:05.120 for your own needs or your business needs. The 00:13:05.120 --> 00:13:07.620 blueprint, as we've discussed, is clearly out 00:13:07.620 --> 00:13:09.740 there now. Thanks for joining us for this deep 00:13:09.740 --> 00:13:12.320 dive into conversational AI architecture. It's 00:13:12.320 --> 00:13:13.980 a fascinating space. Until next time.