H9CEAI · Customer Engagement & AI
Session: Tuesday 14 July 2026 — Full-stack build: chatbot + MCP integrations
Speaker: Victor del Rosal (and class)


[00:00:06] Victor del Rosal: Hello, how are you doing? You're very welcome back. Good to have you in customer engagement and AI for today's session, 14th of July. And as I said to you, I find that it's more valuable and useful to have this session recorded. You can pause at will and engage with it in your own time. Again, for the more technical parts of the course, so we'll continue building with MCP integrations, the chat bot as well, the backbone that we had, so we'll get to that momentarily. So, again, thanks, thanks for joining. So, just a quick... housekeeping for, given that we have been working our way to the, again, this technical part of the course.

[00:00:53] Victor del Rosal: And again, I'm conscious that it slowly builds, you know, the knowledge slowly builds on how to use AI to create not only the front end part of your. of your web applications, of your services, et cetera, but also the back end, and that takes time. So, you know, it's a set of competencies that they build, and you know, you slowly get used to it, especially working on terminals, so your AI on terminal as opposed to the cloud-based environment. So, we've seen this. over the course. So I wanted to delay the CA. So initially I thought we could have the following week or even as early as this week. But we're going to push it to the 28th of

[00:01:36] Victor del Rosal: July, so that's in two weeks time. And this corresponds to CA2, so that's 20%. Altogether you have 30% for your coursework. And again, all the percentages, they remain the same, 70% for your final project. That will be the final project. So we'll talk about in the last couple weeks. We'll get to that. So just a quick reminder of this. You can set it on your calendar. And for this, CA2, you'll have six days to submit. So it'll be due on Monday, the 3rd of August. So there'll be plenty of time to work on that. And you will expect, or you could expect MCP, Chatbot, and other builds. So the backend part of our course we've been working towards.

[00:02:19] Victor del Rosal: So we'll have plenty of examples and practice in upcoming sessions. I'm also planning to have. one-to-one feedback sessions, live sessions in the coming two weeks. So keep an eye on that as well. Okay, so let's get right to it. And just a reminder, we saw previously the model context protocol. And a good way to think about this is a tool, universal connector. of external tools, APIs, and other sources that are other than the model, than the large language model itself. So it's a connector, like a USB universal connector, you know, of sorts, where you can plug in, you know, a number of, again, APIs, tools.

[00:03:06] Victor del Rosal: etc. So I suppose what I recommend, and we're going to see that, use that approach today, is get familiar with using it, integrating into your AI. There's no single way to do it. There's no one right way to do it. But we'll see that, you know, there's an approach that probably will work for you, testing the service first. Instead of building the front end, we're going to start building the back end and integrating these services. So that's what we've done previously. So let's review the, previously we looked at the metal vet care in modern Iris Vet Clinic. So you had access to the Google Sheets,

[00:03:44] Victor del Rosal: the API. So that's exercise from the previous session, 7th of July. If you haven't caught up with that, make sure you do catch up and complete that first. So it'll make more sense. So we're going to build on top of that one, add more integration. So in my case, what I did, and I'll show you the local version first. We have this, this, this service like, like this, like so, you know, and I could say the question I like to ask, you know, what is the most expensive service on the list? Now, I've updated this since we last spoke, so it should come up with something outrageous for dental extraction.

[00:04:31] Victor del Rosal: And it immediately highlights and, you know, gives the red flag, say, look, this is probably wrong. So it's thinking, and that's the whole point of testing this. It's, you know, it's thinking that it is probably wrong. but that's the correct or that's the figure that I entered on the Google spreadsheet. So I've been now a point to say around this is it's important that if you test it yourself, you should get the same answer because that's a live dynamic Google Sheet. And if in my case, the source, if I'm updating that, that's the whole point that you can. catch up to whatever is the live version of it. So never make a copy,

[00:05:12] Victor del Rosal: never store it as a, never ingest it locally or in your code base. So don't hard code it is a point. Make sure it's always pointing to the live endpoint, which is in this case a Google Sheet. So okay, so we test it. It's looking good. So now let's see what it's asking us here. And it says, we want to add the following features, automatically check for Irish public holidays. Are you open Monday or, you know, coming Monday? Check for weather triggered engagement. Is it too hot to walk my dog right now? And then, so we're going to do that first, and then we're going to add other, we're going to simulate a panel of users

[00:05:56] Victor del Rosal: that might interact with this particular service, and then we're going to keep adding features. So just to let you know, by the end, so I'm going to do it here in the exercise, and I'm going to do my own, you know, my own creation. Now the thing is, you probably, when you do it yourself, I'm going to ask you to submit it yourself by next Monday. Do your own thing. Do your own version. It might be the same or a different API altogether. You will more than likely, I expect you to have, and you probably will have, very different features that are requested by the simulated users. So there'd be no two code bases that are the same.

[00:06:39] Victor del Rosal: It's almost impossible to have the exact same features or code rather. So again, make sure it is your own process, your own workflow. So this exercise, what we're doing is we're going. end-to-end in terms of the, you know, not only the front end and the back end, but also the innovative process or the process of iterating and innovating to cater to this clinic. So think about that. I want you to explore a few different ideas. And when you submit your link, again, I expect to see very different features. So this will probably take you a couple hours, is what I recommend, that you play with it. You can use the previous version that you

[00:07:29] Victor del Rosal: had from last week, or you can start over. That's fine. Okay, so let's get started. And what I'll do is just, you know, what I would normally do, you've seen me do it a few times, is you point the AI on terminal. In my case, it's Claude, my own version of it. Well, it's Cloud Opus 4.8, so there's no surprise there. Nowadays, given that there is Fable, I always like to test that I'm working with Opus as opposed to the more expensive or token-hungry Fable, which is not a good idea. So if you're in a subscription with one of them, it could also apply to the latest 5.6 SOL. with ChatGPT, I might recommend to downgrade to the

[00:08:16] Victor del Rosal: next, you know, not as expensive and not as smart, but it does not eat your or deplete your tokens as quick. So that should be working there. Okay, so let's get started. And what we could do is, what I always like to do is... Or you, you know, again, you could do it a number of different ways. You could start by telling it to reacquaint with a project. So in this case, I'm going to go for just telling it, I suppose. giving you a screenshot of this exercise. Look, here's what we're trying to do. Okay, so I'm going to take a screenshot, and I'm going to drag it. So you can't see that on my screen right now, but I've dragged it,

[00:09:01] Victor del Rosal: and it's right there. And I can just say, we're trying to build this, so we're going to start with the public holidays first. So we'll start with that. And then I'll say, so what I need first is a public API that is specifically catered to the Republic of Ireland. In other words, I want to know what are the public holidays. And I want to bake that in and embed that into my application. So if users. ask, you know, are you open next Monday or whatever day of the week? We can confidently say that it's not a public holiday, so we can validate that. So can you look for, do some research for such APIs?

[00:09:54] Victor del Rosal: Okay, so from this point on, again, I'm going to run it real time. Some things might take, you know, a minute or seconds or two, 3 minutes. So it's okay. You can fast forward. So that's the convenience of the recording. But you see right now, it's, so what I did, again, I fed it the actual exercise. It has a context of its code base, so it automatically reacquaints and it is doing that there. I look at the project first, then it's going to research Irish Public Holiday APIs in parallel, and it found one. Apparently, it found this one. Date.nager. at. So we're going to say that that's probably a good one.

[00:10:41] Victor del Rosal: And it's verifying that it is good for the Republic of Ireland. That's good. and starting to see how to incorporate it. My earlier comment on the model, I told you so, it set the model or confirm the model to Opus 4.8. That's the default. Going forward again, I don't want Fable, which is more expensive. Okay, so it's looking at that and wearing that in. So again, this might take a minute or two. Now, from the point of view of the user, once this is baked in, it will probably, the user will probably not notice, right? Because it's just assuming that we know our calendar. So this is more of an internal thing so that we don't schedule something on a

[00:11:31] Victor del Rosal: bank holiday. So again, it won't impact users. It would impact if it was, if we scheduled something and we were not open, right? So in this case, it's more than an invisible thing. for the user. So we'll let this run and it might be a minute or two more. We'll come back to it momentarily. I'm gonna do something right here, which is let's test it before the it's deployed. Okay, that's probably not working. Sorry, that's just the worker, which is an endpoint. Never mind. So let's go ahead and ask it for. What is the public facing URL? Remind me. We can actually do these sort of things, you know, as you're working,

[00:13:42] Victor del Rosal: you can still ask questions. Sometimes it might not answer, it might take longer, but in theory, it should be. To be OK. We'll give it another minute or two. Okay, so let's test the public facing URL it just gave us. This was the same from last week. Should be the same. Okay. And let's test it. Were you open on? Or, or let's let's test it - what what are? Actually, you know what? I'm going to check my calendar before I ask. Okay, so July. For August, it should be the 3rd. Are you open on Monday, 3rd of August? It may or may not have plugged it in yet, so let's test this. We're closed, okay, so it it came it came through,

[00:15:15] Victor del Rosal: that's fine. Okay, so we were too late for the test. So had we not done this... Have we not plugged it in, it would have probably said, oh yeah, you might be open. So again, it's just a precaution. All right, so. That's fine. OK, so let's keep testing. What are what are all bank holidays? In which you are not open. in 2026. Let's see if that works. And by the way, you know, you could argue, oh, look, it already knew, look, we could have tested that before and after, but I suppose it's a level of redundancy. So, you know, as you're getting closer to production, you want to make sure it has these sort

[00:16:09] Victor del Rosal: of, oh, look, it's the service is down. OK, still. Okay, so I might be making changes there. So let's see if we do the same, just out of curiosity, if we put this on the local. Let's try one more one more time locally, so that's a local file as opposed to the file that it's pushed to to the server. Okay, so it's also down, so... Now let's keep trying. So I'm curious now. Let me try the same question. It might have just taken everything offline. All together. We'll see in a second. Yeah, no, it is offline. Okay, that's fine. So we'll be patient. Still working. We'll leave it a few more, couple more minutes.

[00:17:10] Victor del Rosal: Should be done shortly. It looks like something's happening. Give us a... theory of activity and it's supposed to be live, okay? Right. Okay, so second tool, check opening alongside, sir, okay, the... It's A2, right? It's connected to the brain. Okay, so let's do, let's check that in our live interface in the public URL. I'm going to refresh. So I did a refresh there. And let's ask the question. It's thinking. Okay, we're close to New Year's Day, right? So it goes through all the list. Easter Monday. Okie dokie, that's fine. The Monday the 3rd, so that's upcoming. October. Very good. So it's working. So that's good.

[00:19:43] Victor del Rosal: So now we're more confident that we won't be, you know, won't be booking. So that's step one complete. Very good. Excellent. Now, the next one is slightly.. . more entertaining and it's for checking the weather, right? So let's move on to that. So let's go to our terminal. Okay, well done. Let's move on to the weather-triggered events, right? Let's build in the feature where the user can check if it's too hot, for example, to walk their dog. So again, this is not directly connected. to our clinic services, but it's more of a service for users to know if it might be too hot or otherwise. So let's,

[00:20:38] Victor del Rosal: I'm going to assume that you will need, of course, the user's geolocation or they might need to indicate in their, you know, where they're based. So let's think about it for a sec. Let's plan this. And then we need to look up an API that is the most useful for this to work. So what do you suggest? So again, we're going to let it plan. Okay. Now, in some cases, by the way, if it's a very, you know, intense planning or, you know, something that requires a lot of intelligence, you might use a better model. So I might use Fable 5. And then when it's time to build or execute that plan, you switch back to the,

[00:21:24] Victor del Rosal: you know, the not as intelligent model. And the way I like to think about this is in the in the in anthropics ecosystem, Fable Five is like the very, very expensive, very fancy consultant, very specialized. So you pay them, you know, thousands an hour or a day, whatever it might be. Opus, again, very senior. Very intelligent, but you know, you don't pay them as much. But again, they're quite capable, but not as smart, not as, you know, world class necessarily as Fable Five. And Sonnet might be your, you know, your junior employee, etc. Again, at a different rate. So in terms of the token, that's how they're, you know, the pricing,

[00:22:08] Victor del Rosal: that's how it works. Very expensive for Fable, not so expensive for Opus, and the least expensive, but again, very cost effective because you're still getting a lot of intelligence, a lot of bang for your buck. Okay, right, so let's run this and we'll wait a sec for the outcome. Okay, so I wanted to share with you another trick or tip, which is, so first let's read what's happening and then I'll go to that tip. Okay, so it's telling us. Met Aaron, open up with timed out. Okay, the plan, location, 3 tiers. Use my location button in the chat. Makes sense. Or the user just types Galway. Fair enough.

[00:24:28] Victor del Rosal: The real product is in the weather. It's the verdict. Anyone can get a temperature. What a vet clinic can uniquely give this judgment. Go now, caution, avoid, and here's a better. Window. Returns a walk safety call built from feels like temperature. UV humid. Okay, that's fair, very good. So you see, that's the plan. That's the value add of discussing this, right? You're not telling you what to do, you discuss it. That's, I wanted to show you that as well. And it hooks back to the clinic rather than the weather. Okay. Dog walk safety only answers is a tutor with a book. Okay, so it's telling us, do we want to focus on the, you know,

[00:25:16] Victor del Rosal: walking the dog, right? Which is sensible, but also saying all pet outdoor advisor. Same engine also covers rabbits, not really, because, you know, I don't expect people walking their rabbits. So let's stick to dog walking. So #1 is fine. Verify a real source first. I search for the wristbands. Okay, so it's asking for that. That's perfect. And we submit that. Okay, so the tip was in parallel, you can do your own research for APIs, right? So previously I was looking at this and it turned out that there's an API called, have you open Mateo or like meteorologist, meteorologist, have you considered?

[00:26:03] Victor del Rosal: Deep open. Video API. So, we'll leave that as a question to consider while it's doing all the planning. And again, we'll leave it for a second to think. Okay, it turns out that's the one that's recommending open media. That's fair. Fair enough. Very good. Maybe it already is covering this, but I'm going to tell it to add a disclaimer. Add a disclaimer. Legal waiver. Compliant with you, the public line. Laws, et cetera. Okay. All right, so that was a fairly sizable build. The page is live, the weather feature shipped and verified. The API is baked in there. Okay, looks like there's no, there was no sign up required.

[00:37:24] Victor del Rosal: Sometimes you might need to sign up. Just keep that in mind. Now let's, moment of truth, let's see if there's a... Okay, so live service, is it okay? So let's, it's right there. Is it too hard to walk my dog right now? In. Let's assume I am in Dublin. Let's test it for Dublin. I'm in Dublin. Is it too hard to? Right now. Let's see if that, if that will work. Okay, it's currently 21 feels like, so it's water risk, fine for a steady. Okay, fair enough. Now, let's go somewhere else, somewhere very, very hot. Yeah. I'm in the middle of a desert. Is it too hot? The dog. see what it comes up with and how it solves this. Okay, try Sahara.

[00:38:37] Victor del Rosal: and then see if it can continue the conversation. Okay. Okay, assume, okay, that didn't work. Assume it is 40 Celsius. Let's see what it thinks. Okay, the speaking chair is too dangerous for any dog. That's well. Okay, so we tested a few things right there, and one is that it does pick up the, you know, it carries a conversation. Sometimes the chat bots, they will break and they will not. So you're asking something and you ask a different question or you want to continue. Continue the question, but it doesn't go back to the history or the context. So just focus on the question that you just asked. Assuming it's 40 Celsius by itself makes

[00:39:30] Victor del Rosal: no sense. Given that it looked at the context window before it, it does pick that up and it's smart enough to know that or realize that. However, do check that in your own in your own deployment so that people can have a free-flowing conversation. Otherwise, if it stops and breaks every time, say, what are you talking about? You know, who are you? So again, that's something to be mindful of. So that looks fine. So far, looking decent enough. We can do more tasks, etc. Try all the different features of it. So now the next part of the assignment I want you to consider yourself, and we're going to do this right now, is to create a virtual panel of simulated

[00:40:17] Victor del Rosal: users. Okay, so that's the fun part. Where would you do just that and see what they what they might think? So, again, this is, there's no one way to do it correctly, so we'll just chat with the with the, you know, with the AI. I could actually trigger Fable, so I'm just gonna, I'm gonna upgrade. You know, to the better model, so model be able. And it asked me to confirm, yes, okay. So for this part only, we're going to plan. We're in plan mode. We're not in build mode. And it'll use less tokens. So that's important. Right, so we want to plan what other features might be useful for this medal vet for the clinic.

[00:41:05] Victor del Rosal: And the way we want to do this is to plan and simulate users, real world users. Let's assume they mostly live in Dublin, where this clinic is based. Look, I'll let you decide where exactly in Dublin, if that makes sense. We're going to make or simulate a number of users thinking that we have a customer base of let's. Let's think of about 400 users, 400 clients that come through the doors, in and out for the last couple of years. So we want to create, I suppose I want you to plan this simulated marketing research, market research rather. and talk to them, see what they like about our service, what they dislike, and what features we should add.

[00:42:03] Victor del Rosal: And again, we want to add more MCP type features, more API integration based on their advice. So can you plan this market research? Don't run it yet. Just stop short. We're going to switch to the Opus model when you finish your planning, and then we'll execute that. That's it. So we're going to let it run in this more intelligent version. In reality, you might not need this. It's not very, you know, intensive planning or very sensitive type of planning. We could have just stuck with Opus, but let's see, you know, how good it is. And then we're going to run it with Where it switch back to open, so later on from now, OK?

[00:46:03] Victor del Rosal: Okay, here we go. So the plan is written and it has, so it wrote an MD file. Okay, if you look at the... Better Met Care and let's see research plan. Okay, so there it is. Let's have a look. And Simulator Market Research, Metal Vet Care Client Panel. Plan, it has the so it shows Ranela del D6. And it gives us, you know, the so just an idea of the population, the universe of clients, the layers. segments, the interviews that it's going to conduct or simulate, etc. So bias control, seeded candidate list. So in other words, it designed the research, the market research for us or for Opus to execute. So let's do that. So let's switch the model to Opus.

[00:47:08] Victor del Rosal: Less expensive. And then we can go ahead and execute the market research plan. Now I had a command which I created myself. It's called very simply DIY. There's a more elaborate version that I call crank, which again goes through more, I suppose it's more robust, but for now, this is fine. And so it won't ask us any questions. It'll do this end to end. So let's go ahead and do that. And let's see if it... It's triggered. Okay, DIY. So again, it's autonomous end-to-end work by the model. No questions asked. Now the other one, bet weights is, I said before, is what the model bet its weights that the answer is going to give us is correct.

[00:48:02] Victor del Rosal: So just the way to force it a little bit more. And it's going. So this will take a little while because it's going to build the 400 client register first and then add out of panel. So it might spin sub-agents, it might do a few things. I'm gonna ask it to do that. Do feel free to spend, I'm going to say Sonnet, sub-agents, if useful. To speed up, OK. Now, Sonnet is a tier below Opus, so it won't be as expensive as Opus. Never mind, it won't be anywhere near as expensive as Fable. So I'm just giving it that tip if it wants to do that. Noted, I run the panel on sign observations in parallel, building the richer. Okay,

[00:48:50] Victor del Rosal: that's fine. Okay, so let it work. Again, it will take probably... A little while. I've just asked there to resume with whatever data it has at the moment, given that it's looking at 40 depth in-depth interviews and other other sessions. So say, look, that's enough. We'll see how it goes, but again, you can appreciate that you can do a lot of work in parallel. So these are sub-agents, by the way. And every modern AI terminal setup has this idea that you have your main agent, you can spawn, or it can spawn sub-agents doing parallel tasks. So there's parallel processing, that's an important concept as well.

[01:00:51] Victor del Rosal: So it's saying there, it's pulling whatever the panel has produced so far. 52 of 3 agents, or we were almost done, have returned interviews, usability sessions, and it's being extracted right now. So we'll wait for that. Now, in the meantime, you can see that the research is being modified real time. So There is information that is landing there as we speak. So again, all of this is being written to our local folders. and it's being updated. Okay, so that's important to note. There's a folder there, and we can check in and inspect what's happening. And again, all of that based upon from the plan that

[01:01:37] Victor del Rosal: was written by the more capable model, Fable. Again, that was probably not necessary for this example, but it demonstrates that. As well. interviews are landing, etc. Panel data is in and it's rich. The decoys were rejected, 36 out of 40. The endorsements are trustworthy and only 5 of 12 said they used the bot again. Okay. So that's good, that's good. So pulling the verbatim failures, live sessions, found real defects in our ship bot. Most damning, the sheet has broken prices, yeah, and again, that's our fault, we did it for other test purposes, and the bot defends them under challenge, no typo, that's the life price because of a rule I

[01:02:24] Victor del Rosal: wrote, okay, right in the front, so it's writing those findings, okay, now. Let's see, that's yeah, so the findings are here. Very good. Decode. OK. Cost estimator vaccine forming due date. OK. All right. So what we want to do is. is the following. So we're going to ask it. I want to ask it once it's finished. And again, we're doing this before it's done, rather. So it'll be sequential, not parallel, which is I'm going to ask it to basically based on the output of this, tell us what features we should build, okay? Right. So we want to focus on one or two features that we should build. So right now you can finish your finish

[01:03:25] Victor del Rosal: or wrap up the research process, extracting all the insights. And what we want to know right now is what feature one or two features are the ones that we must build for this chat bot service to work and work best based on user expectations and the UX on everything that was said in these panels, et cetera, usability. So just tell us, what should we build? So there we go. Now, it's not finished the previous process, so it's still wrapping up that. So again, it'll take another minute or two. Now, what I just showed you right here is a typical, these are the typical wait times, right? Previously, in other sessions, just for this very reason,

[01:04:12] Victor del Rosal: I have not run them. So I suppose it is important that I run them. in real, in their actual time that they take to process. So you have an idea how much, I mean, you probably have your own time estimates after working for a few weeks, months, or probably years if you've been doing this kind of thing for longer. But again, it's just a good idea, instead of me editing this, just to show you that you know what it actually takes while you're running this. Now, the way it works, I suppose, just to add a footnote to that is, if you're doing this kind of thing for a number of other, you know, projects,

[01:04:53] Victor del Rosal: work projects, personal projects, you probably have a few terminals running. at the same time. So that's typical. You have a two, three, 4, 5, so your brain can get fried, your own human brain, but you have, you know, you're supervising a number of agents for this reason, because you don't want to be waiting for a single agent to do a single thing. And you're, you know, you're just there doing nothing. So again, that's what you might expect. Of course, you know, you have the demand for it or you want to do it, you want to test a few things in parallel, if that makes sense. Okay, so it'll be a few more minutes and then, so what's it saying here,

[01:05:35] Victor del Rosal: wiring the new tools in? So it went ahead and wired in. I suppose it has the... Okay, so it says paddle complete, 54 agents, independent red team killed two things, and confirm the rest. Okay. Right, so that's finished. We also added in a request. and deploying and testing the exact failures the panel found. Okay, so it's going ahead and doing that. So again, we'll give it a little bit more time. All right. Okay, so here we go. So it says build 2. Cost estimator is 1, reminders, both are now live. The cost estimator, 29 out of 40 picked it, the loudest signal in the panel. This is not really a feature request,

[01:08:16] Victor del Rosal: it's a complaint. Every last client in the panel left overbuild shock, not care quality. Okay, so that's good. So no bill shock. Vaccination, warming, reminders. Tell us everything does already does manually, a note on the phone. So just a, I suppose, a reminder. Okay, so symptom, okay, so let's. Let's focus on one for the moment. Now, for the second one, the vaccination reminders, what would follow for this particular example is you would need a system to log users, remember them. So that's a Google authorization integration. We won't do it today. Just so to focus on the other one. So the cost estimator. So

[01:09:05] Victor del Rosal: Let's do that. Let's implement number one, cost estimator only, and leave all other requests out of scope for now. So go ahead and implement a cost estimator, please. That's it. OK, so that will be our last build for for this. for this particular project. So in your case, I want you to do something similar. Now, again, feel free to do, and you know, I will request that you do the, you know, the first two features, the public holiday, Irish public holiday check. Weather trigger, trigger rather alerts, that's fine, or you know, the should I walk the dog, etcetera. Make sure that works for you. You plug in this, you can do the exact same MCP API

[01:09:58] Victor del Rosal: integration that I used for this for this session, and if not, it'll be in the notes, so you can ask the. the chat bot and it'll tell you exactly which one I use. Now for this one, for the virtual PAL simulated users, feel free to implement any approach that suits you. You might say, look, I don't want to interview 400 people. I just want to talk to, let's say, a panel of 12. And these will represent, will be representative of the whole. your clientele, our client base, you know, and work with that. Or you might say, someone else might say, you know, actually, I want to do market research for a real, you know, more real world market research.

[01:10:41] Victor del Rosal: Get, you could use, for example, other tools. You could use standard Google search, etc. Or you could use in cloud-based, you know, Claude or ChatGPT or even Gemini, etc. There's an option for deep research and do that and then export that as a markdown file or as a PDF. and feed it and say, look, have a look here, what was our clients have thought if you want? And then you can choose. So there's a few different ways to do this, right? To integrate real world research, you might just have, if this were a real, you know, vet clinic. you would have that from requests that people tell you, look, we're sick of this,

[01:11:30] Victor del Rosal: or you have complaints, so you build it. You have enough complaints or requests. So that's, you know, but I want you to simulate that. So I'm eager to see what you're planning to build that is your own flair, your own thing, and again, it'll be. Everyone will have a different code base and different flavor of things, right? So in my case, I've deployed it not to a GitHub page, but it's through Cloudflare. It's up to you. So this is just my own, sorry, no, it is through GitHub. I apologize. It is through GitHub. It just shows my domain name, Victor Vyas. Which, by the way, if I were to, just really quick,

[01:12:10] Victor del Rosal: if I were to change this to, let's see, GitHub.io, I think that's the right construction. it works. So it just redirects to my domain name. So it is GitHub. Okay, so do the same right now. You can also test this again locally first without having to go to GitHub. But again, the GitHub part should be now trivial. You can even automate that. And you can ask your, if you haven't done it, you can ask your AI to do so, to automate the login process. It's just, you're pre-signed in to GitHub. And all you say is, look, commit this to a GitHub page, and that's it. It'll take care of everything for you, as long as you're given the right

[01:12:55] Victor del Rosal: authorization. You might have to still hit accept. along the way, but that's better than doing manually. So again, I expect you to empower your AI agent and abstract out a few of these things. Okay. So it will be a building and it'll be, I suppose, finished. And what I'll do is make sure that the link is up, well, you have this link and you'll have it here in this environment in the context. You'll be able to see it for yourself, etc. So keep an eye on that. And, as we, as we, as we finish, so saying it's deploying and testing, including the case the old version got suddenly wrong. The tool works,

[01:13:39] Victor del Rosal: but the pics are wrong. It chooses rabies vaccine as a puppy. So, again, it's doing its own testing, right? So, it's again, this would be very important. And in the real world, I would not recommend that you release something until it's properly tested, embedded internally. So just be mindful of that. And let's see, she has exactly right services. OK, rewriting the bundles again. OK. That's fine. All right. Okay, so while we're waiting on that, just a quick reminder of the, and we're going to finish with this reminder of the CA2 in a couple of weeks' time. So that's the 20th of July, 20%. You can expect a similar exercise or a number of exercises like this one.

[01:14:29] Victor del Rosal: that you will build over, you have a six day period or window to submit due on the 3rd of August. And that's it. So we will, as I said, we will have at least one live session where you can ask queries, individual, you know, one-to-one. Also, as we work our way to the final project, start thinking of ideas for your final project. So these will be your own ideas. I'll give you ways to. probably generate some of them if you're unsure what to work on. But again, what I prefer, and this is from other cohorts from experience, is you come up with your own use cases that mean something to you. You know, it could be literally a pet project,

[01:15:15] Victor del Rosal: if you pardon the pun. It could be, you know, something. nonprofit, for volunteer work. It could be your own startup idea or something adjacent to that idea. It could be a tool for work that you might want to build internally for your own purposes. Again, you know, look at the compliance of it, the data governance, all of that, you know, so it is careful, etc. So Now, keep in mind also that speaking of that corporate angle, some of these would not be probably aligned with, you know, your real world work for a number of reasons, for, you know, the data compliance, GDPR, et cetera. So maybe the safer route is other types of projects. So, but again, anything.

[01:16:01] Victor del Rosal: will be, we will explore and we'll have a dedicated session to work on that and one-to-one time, you know, enough time for queries ahead of the CA2 and afterwards for the final project. So that'll be the, that's the plan for the upcoming weeks. So we'll leave this running and we'll see, I suppose you can come back, you can fast forward and see the outcome. So we'll just give it a few more moments and then we'll wrap up. Okay, so it looks like it's finished. And there it is. Cost estimator is built, deployed and live. Everything else untouched, okay. Ignore, especially pick the wrong services. Okay, so let's give it a go.

[01:19:13] Victor del Rosal: And open the... All right. and ask it to estimate. Can you estimate the price? Or a number of. A puppy or puppy services. Give me the best posts. Let's see how this works. Okay, so there it is. What will a new puppy? That was a better phrasing right there. Right in front of me. How much for my cat's annual checkup? Is it? Okay, very good. So there it is. Okay, so estimated cost, new puppy, first year. There it is. That's useful actually, right? So you're new to the whole thing. And again, if you click on this, you have an estimate, right? Very, very useful. So again, just as a way to wrap up and show that

[01:20:12] Victor del Rosal: based on our Google Sheet that we had, fairly simple, straightforward, we can build a number of services that are relevant to that data. It's dynamic, we can, you know, edit, change prices, etc., as long as they're reasonable, not like the ones that I changed, and build something interesting on top of that. So, keep iterating in this again, this whole approach that I just demonstrated now, including how long it takes. is part of what I want you to keep practicing and keep working on over the coming weeks for CA2 and for your final project. So this is what I expect you to be building and deploying. Again,

[01:20:55] Victor del Rosal: as I like to say, you're dangerous in a good way with these tools. So build responsibly, build sensibly. You have now not only the front end and the back end, And you also have tools, real live access, you know, how to embed data, real data, dynamic, et cetera. So again, very, very powerful tools that you're working with. And I'm delighted to see that you're doing such great work and submitting great work. So keep at it. Now, one last thing I'll say is, it is normal to feel overwhelmed if things don't work out, frustrated, you know, going back and forth. So if you get errors, feed them back into the panel, say, look,

[01:21:38] Victor del Rosal: here's what I'm getting, take screenshots, you can drop in images. So it's normal. And you know, you might get stuck, so come back later. So give yourself time, but do keep tinkering and playing and testing a number of times. So it's not a one shot kind of thing. You do it over and over. Okay. So listen, that's it. Thank you very much for your time, for your patience. And I'll talk to you very soon. We'll be in touch. All the best. Take care. Bye bye.