Data Analyst Portfolio Project with AI | Supply Chain Analytics

Google just launched a brand new AI agent called Spark, and today I am going to have it complete end-to-end data analyst project. I'm gonna have it source a high-value project and have it clean it, define the requirements, visualize it, and even produce a one-pager at the end or some sort of portfolio. Now, if this is your first time here, I'm Kedisha. I am a self-taught data scientist and instructor for over 5+ years. I have been coaching career changers to break into data analytics without going back to school and wasting more money on courses that just don't work. And if you ever wanna work with me and get one-on-one coaching to break into data analytics as fast as possible. Click the link in the description if you want to learn more. All right, so let's get into it. So we're going to be using Gemini Spark to build an end-to-end data analyst project. The first thing I wanted to do is to find a high-value data analyst project, right? We're not just going to find something on Kaggle or whatever. We want to find a high-value data analyst project. But before I start there, what exactly defines a high-value project? So there's always 100 things that you can do in a business There's going to be certain things that are the biggest bottlenecks, or the certain things that affect the bottom line the most. So what I want to do is I'm going to pick a target industry, and I'm going to ask Gemini to basically source out a dataset and a project for that. So I actually did a video not too long ago. I used AI to build a $100K per year portfolio, and I used Manus to basically create a project. And so what I'm going to do is I'm going to take the prompt that I did for that one and paste it here. And so what I'm saying is I'm an aspiring data analyst. I want to build a portfolio project on a high-value problem in my target domain, supply chain. Can you find me real datasets online and build some CSV files or file for my data analysis project? Excel, Tableau. I'm also going to say I want real datasets Not on Kaggle. All right, and then we're just going to press enter and see what's happening. All right, what I do like is just like Manus, it's showing exactly what— so I've identified potential supply chain problem areas, potential data acquisition. I've begun brainstorming potential supply chain problems suitable for a data analysis project. It's showing you what it's doing so you can QA it real time. When we're doing anything with AI agents, Right? We do not want to just set it and forget it. You want to see what it's doing. Let's say it says USAID shipment data, right? Let's say for whatever reason, I'm like, no, I don't want it. I don't want that. You can literally— I can literally talk to it in real time and say, no, no, no, do something else. Hey, I've determined I can't directly download that dataset. So, okay, now it's going to explore other methods. Okay, so it looks like it's doing some work, like data cleaning already. I don't want it to do that. Do not do any data cleaning on it. I'm going to stop the response. Okay, to ask new questions, you can wait until the response is canceled or start a new chat. Okay, well, that sucks. Okay, well, that sucks. Well, man, you can just talk to it in real time. All right, so it's already doing some data cleaning and stuff like that already. We'll just wait for it and see what it's doing. Okay, look what it did. It provided the dataset, it provided the schema. So we got products, geography, shipments, vendors. This is good because I was actually going to ask it to create schema design. It's given us fact table shipments. This table contains transaction-level shipping records. It has a dimension table products and other dimension table vendors. This is good. Okay, so, so far let's talk about what I like. It obviously provided some data. It went ahead and even did a schema design. It looks like it got data from the USAID Global Health and Supply Chain Program. The data has been structured into a star scheme, which is great. Now the only thing is I wish it didn't go too far ahead and, you know, started exploring the dataset or just coming up with these questions. What are on-time delivery, things like that? I would much rather we take a step back and say, hey, I found this data and just provide like a project overview first. Before anything else. So what I want to do— okay, we went a bit too far ahead into data exploration. I want to start with a project overview, the business problem I am solving for this, you know, portfolio project, and 5 to 10 You know, key questions about it. Let's go from there. Let's, let's take a step back. It went a bit too far ahead. And what I don't like is like, I can't like just talk to it while it's working. With Manus though, you can talk to it while it's in mid-prompt and it can just like redirect its work. All right. So let's see what it's doing here. Here's a structured framework for your portfolio project. Prioritize business impact, operational metrics, data storytelling. All right. Global health supply chain and logistics optimization. Target domain supply chain. Case study is the USAID Bureau of Global Health. This is really good. So this is like real data. Okay. So we're going to focus on end-to-end logistics performance. And we're going to use historical data across 20 nations. This is really good. So the USAID Global Health Supply Chain Program, you know, it procures and delivers life-saving health commodities for like treatments for malaria, HIV test kits. And supports countries worldwide. So this is some big time stuff, you guys. So this is big time right here. All right. Business problem, right? Business problem is a big thing. There has to be always a business problem. There always can be 100 different angles you can do when it comes down to analyzing data. You want to just choose like one. So you're not just like focusing on one project forever. Right? So hiring managers in supply chain look for analysts who understand how operational performance affects financial bottom lines. Absolutely. Bottlenecks. The most important analysis that anyone can do in any domain. Is bottleneck analysis. 100%. When you first sign on to a company, heck, even in your interviews, always relay some sort of expertise or whatever when it comes to bottleneck analysis. That is huge. This is how you do, again, high-value projects, ones that attack the bottlenecks. All right, so logistic SLAs. Delivering life-saving health commodities late is not just a failure of service level agreement, it directly impacts patient health. Number one, right? So we must identify which countries, shipment modes, product categories are experiencing persistent delivery delays and Another challenge is financial leaks, cost containment, international freight shipping costs. Okay, so what I want to do is I just want to focus on one. So this is one thing, right? When it comes down to we're making this like a portfolio project, it's good to just focus on one. So it's good to have a project that relates to some sort of bottleneck analysis because again, I think that's a, that's a $100 million type of analysis right there. Bottleneck analysis. Anyways, so what I'm going to say is, okay, let's focus this project On operational bottlenecks only. Now, one thing with agents though, like once you like set things up or set the instruction and stuff like that, then you can have things on like automated tasks and stuff like that. When you first start, like you have to like pretty much set the instructions and stuff. We're going to frame it for operational bottleneck analysis. Now I highly, highly encourage you if you guys do something like this, if actually you should be doing something like this when it comes down to like building your projects. There's always a hundred different ways, angles you can do when taking project for data analysis, definitely, you know, have it look at, hey, what are the different types of angles or business problems that I can attack and just have it provide you, hey, these are different options and just say, okay, I want to choose this one. So we're not doing like, you know, 20 questions that's covering like a whole bunch of different things. All right. So this is going to be very focused on supply chain bottlenecks and delivery failures. Already when I'm looking at this, what I want to do as well is I know in my, in my final presentation that I'm going to have, we're going to create like a final one pager. To put on like our portfolio, I want to have a visual of that bottleneck analysis. Like okay, like this is the top portion, the second portion, and then like a visual representation of where that bottleneck is. Okay, so key operational questions: one, two, three, four, five. All right, so we got six key questions here. And portfolio outcomes for this one—it sucks about like SQL and stuff like that. I don't think there's anything great about or you know, as far as a portfolio project, just having the AI do SQLs. I don't think it's going to be super helpful for you because, for example, if you just. the AI do very complex CTEs or whatever, and you don't even know how to do that, putting that on your portfolio, it's going to shoot you in the foot. I'm going to say, okay, I don't find just having you do SQL very useful. So we are just going to do this regular analysis, visuals, etc. Let's start answering each question one by one. And let's just do 5 questions. I'm also going to have it visualize the table, the schema as well. We're going to have that on there. Let me see if I can scroll up and find it. All right, cool. Okay. And also I just realized like anytime these AIs, whether it's ChatGPT, Manus, Claude, whenever they're analyzing data on the backend, it is pretty much Python for the most part. I think at one time Claude would do stuff with like JavaScript, which was, I feel like, ridiculous. Let's ask it. Hey, let's take a step back. Is there any data quality issues with our data? Let's, let's take a step back and ask if there's any data quality issues first. And again, it gives us our datasets. We have 4. Actually, let's just look at our data really— oh. Oh, look at that. So it's just generating visuals and stuff like that as things go. Like everything— wow, this is fantastic. Okay, well, I want to look at these data. So this is our shipment facts. Look, this is some real good stuff, as we can see here. And I could already see some data quality. So for example, freight cost USD. So again, if you put this, like if we uploaded this in Python or whatever, the data type for freight cost USD, we would expect it to be a A number, but what is it? It's going to be not a number because we got some text in there. So that's interesting. So we got some text in here as well for for weight kilograms. So these are things that I'm noticing. Let's see. Let's let's see how I asked it. Let's take a step by any data quality issues. Yes, conducting a data quality profiling audit. First critical step. Yep. Okay. Non-numerical values weight kilograms as we just briefly looked at. Approximately 3.6% of the dataset. So yeah, see, these are text. It can do some wrangling, replace weight captured separately with a null value, NaN. So that's good. Fixed data types and string and freight cost that we just briefly saw. So about almost 20% of the dataset in here, treating these rows as separate logistics scenario, missing delivery milestone, about 3%, missing optional metrics. These are optional. Yeah, so, and this is where you, as let's say if you want, if this is your expertise is where you want to develop, this is where you can figure out or where you can say, hmm, okay, we got some text values and mixed data types in here. It's giving recommendations for this one, for mixed data types and freight costs. I do, I want to do that. I don't want to just say like null or whatever it is because this must be like a different scenario that has like a different impact or different type of case that deserves its own analysis, right? So me deciding for non-numeric values of weight kilograms, I'm gonna say replace it with NA because it's just kilograms. I'm gonna say for mixed data types and strings and freight cost, I don't want to do a null. What I want to do is treat these rows as separate logistics scenario because these are freight costs. The fact that it says like freight included a commodity cost. Okay, so again, that is a separate scenario. I don't want to say no there. There's a commodity cost somewhere in our dataset. Says commodity cost. Yeah, I don't even know what that means because I don't even see commodity costs as, as a column. Let me go to products, see if that's in here. No, pack price. What I'm going to do is I'm going to treat it as a separate case because where is commodity cost? Where does that mean? Like in real world, sometimes there's more data that we need to go get. Maybe it's in a different table or whatever. And look at that CDN. Yeah, I'm gonna mention that. I'm gonna say treat these rows as separate logistics scenario. These are obviously not something that we'd want to say, and like, this is, this is something we want to treat for later. Okay, so missing delivery milestones. Let's see, for delivery milestones, I want to say these rows should be analyzed differently. But I also want to say let's add another column called delivered, and with the value either— it's going to be binary, it's either going to be yes or no, 1, 0, either 1 or 0. 1 being it was delivered and has a delivery recorded, 0 it has no delivery recorded. That's like another case 1 type of thing. Missing operational metrics, I mean if they're optional is what it is. Structural nulls, that's good. Again, we can't just let AI just do whatever. We saw the recommendations, we looked at the data ourselves, and we came up with what to do, right? I'm going to have it complete the first question again. But so far, I'm, you know, really liking it. Don't like that I can't like talk to it mid, you know, mid-prompt. But, you know, it creates a folder as it builds images and stuff like that. It has them all in here. So this is really good so far. Just another note, whenever you're using AI as your copilot, whatever, you got to learn how to be More of a manager. Okay, so it looks like it's used SQL to do the data cleaning. Okay, great. Let's document the data quality check and solution we did for our end portfolio one pager we will use later. All right, so let's just document what we did a bit quickly so we could kind of save it. This is real data, not Kaggle data. We already looked at it, identified some data cleaning stuff. We use AI to outline or kind of spot check stuff. We also visualize it really quickly and kind of saw some things ourselves. And, but we ultimately decided like what to do, how we're going to implement that. We create a new category. So we did some feature engineering and now it's going to create something in Google Docs in our same folder, which is really cool. Supply chain data quality audit, including ledger. Come on. This is nice. So for the one pager, I would have like, like a screenshot of this on there. For our portfolio because it has our solution and everything. Okay, great. So that's good. Now go ahead and answer question one. All right. So now go ahead and answer question one with our new cleaned. Actually, okay. Now with an updated dataset after the cleaning and some feature engineering, is there any new or up? Dates to the 5 questions we should implement. Yep. So we created some new features, deliver freight cost category. Yeah. Let's, let's start answering the questions. Yes. Let's just go start answering the questions. All right. So in transit, about 3%. Completed, 96% on time. 721 late shipments, 246. Okay. So our global on-time delivery rate is 74%. Target SLA is 85%. So we are below target by 10%. And it created a visualization. Let's look at what it did for question number 1. So I hate pie charts with a passion. So I'm going to have it make this into a bar chart. So it says the program is failing to meet its 85% SLA by about 10%. So 1 in 4, every 4 shipments is delayed. So that's bad. Local distribution centers must carry higher buffer stocks in to insulate clinics from this. You know, late delivery risk, which is true. So now because we are behind or because we are, you know, 10% down below our target, what's the effect? Now these local centers, they should be having a higher stock, basically almost like overstocking to avoid stuff like this. So we're going to just answer all the questions and then we're going to go back and tidy up all of our visuals, right? This is just exploratory data analysis right now. So it doesn't have to be perfect yet. But once we are good to go with everything. Then we will update our charts. I think what I'd like to do is have a donut chart highlight the delayed. So right now it's 25% being late is really good. So 25% of all of our deliveries are late. So that I would highlight, but I'd keep the 74% gray. Yeah, I'll do something like that. All right. Question 2, which countries are experiencing the worst historical delays and where are the active overdue backlogs concentrated? So now we're breaking down our on-time delivery rate. By countries. Kenya, only 59%. Average delay, 23 days. This is awful. What the heck? So now we can break things down by country and then we could also break things down by region if we wanted to. Yeah. Now, when we talk about delays too, like, and this is just thinking, right? The delays could be multiple things. There can be logistical stuff outside of our control. That's one thing to consider. Right? There can be logistical things outside of our control. It's good to have this data here in front of like decision makers so they can pretty much know, okay, this is the bottleneck. What the heck can we do to alleviate this bottleneck ASAP? That's the thing with a bottleneck analysis. It's like, it doesn't matter. Like you have your bottom line or whatever it is, your, your main KPI. But if you have this huge constraint, it doesn't matter a lot. You can be trying to fix all these other things that's not really moving the needle that much. But this, this little, like, you twist this knob just a little bit, this lever just a little bit, and so much, all that bottom line stuff is just like, it'll get like so much better. This is why bottleneck analysis is, in my opinion, the most important analysis anyone can do. Okay. Looks like it created a visual here. Plot 2. Okay. I mean, just looking at it, just off the rip, to me, a lot of clutter. And then we have a dual chart here, which, what is this? Average delay of late days. Okay, so I always like to avoid dual bar charts. So on-time delivery rates. I do like this because benchmarks are a thing. I do like this average delay of days thing. And let's see, the blue is on-time delivery rate. And then, so my thing is this could either be 2 charts, but like I'm mostly interested in places that have higher average delay of days. So I think what I would end up doing is maybe turning this into maybe 2 charts or whatever. Focus on, again, the Pareto principle, 80/20, and make a new bar chart that's just average delay of late shipment days. And we just focus on like the top 5, and then we'll have this here. That's just me thinking. We'll get to that after we finish. Let's move on to the next one. So as you can see, we're still thinking. We're not just letting AI just do whatever. And the great thing about doing it this way is like, let's just say, for example, you want to put this on your portfolio. The biggest thing, or one of the reasons why, like, even for my program, I, for projects you put on your portfolio, whatever, I don't believe in walkthrough tutorials. Hey, like, this is what I'm doing. Like, not, not for like your portfolio, because you need to understand everything about it. My per— my portfolio, for example, I have projects that I did In 2021 that I can still talk through, um, like I did it yesterday, just because I had, I spent so much time on them. And something like this, again, as we're like actually going through double checking, we're deciding like, this is something where like, oh, you can like talk through it like naturally, you know, because you're, you're, you're pretty much learning as long as you go. And you're not just like having AI just like, oh, find a project, produce it, just have it answer all the questions, make all the visuals. And while you're just out doing, because you would not be able to explain. Anything about it. Ocean. While ocean freight is the cheapest, the standard deviation of 21.9 days represents a major operational planning risk. If a clinic expects an ocean shipment of anti-malaria in 85 days, but it's hit with a 0.1, you know, at risk of running out of stock entirely. So clinics must buffer ocean-bound stock of at least 45 days. The Pearson correlation is practically zero across all individual modes. Discovering a lack of correlation is highly valuable. This proves that the shipment size and weight do not drive delivery speed or delays. So a 500 kilogram shipment clears customs and arrives just as fast or as slow as a 50 kilogram air shipment. Therefore, logistics planners should avoid splitting larger orders into smaller batches to speed them up. That's pretty good insight. And one thing I do like is like, it's explaining things. So you're developing some more domain expertise and things like that. And again, these are things that you can further, like read into, et cetera, et cetera. And this is how you develop. Imagine this is like a case study, same type of case study that you do in MBA programs that you spend, you know, $100K to do. My MBA degree. And I'm telling you, this is literally like the same thing, but even way better, to be honest. Okay. So that's an interesting insight. It just went ahead and did that. So there is no correlation with lead time and weight. Now, if we look at it. All right. So ocean shipments take longer, but air takes a lot less. So it's interesting. Any shipments through the boat takes much longer. So that's interesting. Obviously, if we're talking about, this is a US type of thing. So pretty much ocean and air is going to be like, everyone's going to do that. Now, my thing would be interested to know the cost of like an ocean shipment. versus air. Remember, we're talking about bottleneck analysis. We are below our target for on-time deliveries, or our target is like 85 and we're at 74. So that's a big issue right here. And now we're seeing we have 4 modes of transportation and we're seeing pretty much the time. So really, I mean, even though ocean takes the longest, there's less anomalies there, right? So for air and truck, that's The median is, is more definitely lower, but there are a lot more anomalies and variation. Well, not a lot more anomalies as far as like ones that's taken much longer. You know, interested to see what's the difference between air charter and air. There's only so much when it comes down to supply chain. There's so many things that's out of your control. It's in other people's control that you have to rely on other, you know, people. Like if we're behind on certain things, we have an SLA, but you know, we have vendors that, you know, we have to rely on to get us stuff. Like there's so many things, but you can plan. That's what, that's why planning a supply chain, listen, y'all want to be doing data, whatnot. Planning a supply chain is, is huge. A huge thing for like, in terms of a domain to get into. But this is like planning 101. So for example, it's like as a planner, you can see if you can negotiate or get stuff set up to where these distribution centers can have stock. That can basically withstand our lack of being on time, so to speak, right? Because lives are on the line here. You know, this ain't like a, you know, Netflix type of analysis. No, lives are on the line here when it comes down to these KPIs not being met. All right. So let's move on to question 4. Then after we finish these, we're going to look back at all of our charts and just clean them up. All right. Do air shipments to Exhibit Better delivery reliability than surface shipments. All right, historical. So we do way more deliveries through the surface trucks and whatnot than on air. Well, trucks and technically ocean deliveries, because that's where the surface— our on-time delivery rate is about the same. Average delay is about the same though. Interesting. Looking at our pipeline backlog. Okay. All right. Let's look at this Transit speed versus reliability illusion. All right, so global health logistics organizations often pay substantial cost premium to ship cargo via air freight, assuming it provides a more reliable delivery guarantee. However, our data reveals an operational paradox. While air is 20 days faster on average than surface transit, it is no more reliable. Air shipments have a slightly lower On-time delivery rate than surface and suffer an identical average delay. So that's interesting. Like we were just looking at some, one of the other charts that made it seem like ocean was like way too long, but upstream bottlenecks. This indicates that logistics delays are not occurring during the actual physical transit. Rather, they are driven by upstream bottlenecks such as document preparations, custom clearances, import restrictions, terminal warehouse delays affecting air and surface shipments. So I was just telling you guys earlier, when it comes down to supply chain, there are so many different, we don't control anything. So like I said, when it comes down to us being late, right, we have vendors, we have other things. So there's only so many things. So that's why with, when it comes down to something like this, right, if we, we can't really control customs, like document preparation, okay, we can probably get on, you know, what are our workers doing? But at the end of the day, at least just so far when I'm doing this stuff, we're on time about 10%. Down, you know, when it comes down to our distribution centers having what they need, right? It's probably easier or not even easier, but like we could probably control just maybe increasing or sending a larger amount of the supplies so that these distribution centers can sort of have like that stock, like already have that stuff in stock so that, okay, in case something's late, that's not going to mess up providing medicine to our local people. That at the end of the day, that's really what matters. So this is interesting. Let's look at this visual. So as we see now, like air and surface is about the same. This tells us again, it's not a lot of the stuff with the delays is not due to the transport itself. So that's pretty good. All right. So yes, let's move on to question 5. And then we are getting close to cleaning up our charts. One thing I really do love about Gemini, it does show you exactly what it's doing. And it's putting all of this in our folder. Okay, circle delivery. Okay, so this is just direct drop versus regional warehouse drop, and everything's about the same. Uh, let's see, volatility versus speed trade-off. Our analysis reveals that direct drop shipments from international vendors are on average 3 days faster than regional distribution centers. However, direct drop exhibits a significantly wider lead time spread. Value predictability. Regional distribution centers produce a tighter and more predictable Bell curve of lead times. All right, in supply chain management, lead time predictability is often more valuable than raw speed. That is true. Unpredictable arrival windows force local health ministries to carry larger safety stocks and resort to high-cost emergency orders. That is true. Nearly equal division of backlogs. Okay, so the project is complete. You have now completed structural business strategy, data hygiene, data analysis, and everything is still in our. Folder here. Okay. So what I want to do is I want to go back. It says, good luck presenting this to higher managers. Okay, cool. Okay. So let's start cleaning up our visuals for our one-pager. I want to build consulting style BCG and McKinsey images. Let's redo all 5. All right. So now we're just going to clean up our images. We're going to start getting everything going. And now in real time, you guys, like, if this was like a real project that I would want to do, and I would like want my students to do, we would, we would definitely be spending like more than like an hour or something on stuff like this. As we were going, we would spend more time, like really diving in even more and making sure, because again, you're learning some domain expertise. I spent $100,000. The idea of like non-business people. Can just go through MBA program in a year or two, and then just become these consultants doing like consulting and supply chain and finance blah blah blah. You sure as heck can do it too. You can learn to develop that. Like this is exactly teaching you how to develop domain expertise in whatever field that you want to do. And so with these, I would spend more time like digging into you know like say on time, and then you can even ask it like. Gemini, ask it to provide you like, hey, are there Harvard Business Review case studies or other case studies or even YouTube videos that, you know, you could provide that I can, you know, get more insight on certain stuff like this? So, all right, so we're just going to clean up our images, our 5 images. There is one that I mentioned that I most likely want to split into 2, and it is— looks like they are done with the plots, everything's been redesigned. Ooh, look at that. You see that? Let's look at the before and after, you guys. Before, after. This was my idea too. Um, but now here's my thing. Historical baseline is failing 85%. So now we have a, an action subtitle, but here's the thing, 76% on time. What I want to focus on is the delayed. So instead of the, in the middle, 76%, I want what were delayed for the first one. Let's have the bold callout in the middle. And you can talk to— actually, it'll be much faster if you can actually like talk, use your microphone and talk. I'm just for the sake of the video going to type it out. Let's have a bold callout at the middle of the pie chart be the delayed stats. So let's just do that first. And remember, the idea is the bottleneck. So we want to focus on like what's wrong. For that first one. Okay. It's also providing. Okay. This first link is, it keeps being wrong. That's going to be one of the new ones. Uh, but, but, but, but let's see if it's in here. Well, okay. It's not cool. Um, let's see if we can go back to the— okay. So this is a new one. All right. So before we get to, I'm going to come back and just like remove this thing here. It's like maybe make this gray. Or something. Make the blue gray. Gray. Make the blue in the donut. Alright, so let's move on to the second one. Okay, on-time delivery by destination. It's pretty cool. What I want to do is— what are the ones that we— what does this thing about? Okay, replace the stair. Okay. Replace the clutter dual access with a clean horizontal. Sort from Worst performance, the best, the worst performing bottlenecks. Kenya, Zambia. I'm not sure why these they have these red. If all right, I guess if we identify ones that are seventy percent below. But for the OTD severity chart, let's only include the bottom one, two, three, four, five, six. The bottom. 6 performing countries because it's like the 80/20 Pareto principle. For those who don't know, 80/20, you know, 80% of your results come from about 20% of your data points. So we just want to focus on the 80/20. So we don't need to see like all of these. We just really want to focus on these again, like the 59%, 62%, super huge. Like we want to, that is a store. That's the focus rather than like Ghana. Ghana is like. These are like the biggest concerns in terms of the bottleneck. All right, for, let's do the third one. And just a reminder, let's go back to, remember this was the before, this is the after. Much better looking, right? Much better looking. All right, lead time transit distributions by ship mode. I actually like this. We just need to make sure that this subtitle gets some space. Okay, for lead time. Make sure to add some space between header, subheader. They are overlapping. Okay. So this is going all right. So let's look at 4. I actually like this. I would say for this one, I just rather just separate them. This is on like PowerPoint, for example. What I would do is I would just like make these like smaller, just more space. We don't even need the Y axis. So what I'm going to do is like, we don't even need the Y axis there. So that's just clutter. A great start though, as we can see that it's done, but we got to come in and clean things up. What we're going to do is we're going to remove the, the Y axis for both. And we're just going to add a bit more space in between the two. Remove the Y axis or Cs for both charts and add some space between the charts. And there's only 2 bars on each of them, so there's really no need to be like super Too far, but I like I like that. All right, so the fifth one, and let's see if we can find that plot four. That was five. Is it four? Okay, this was four. So let's look at the before. So this is before. This is after. I mean, this wasn't that bad before. Okay, now for let's look at plot five. Let's see what we want to do. So for fulfillment category, let's see what it looks like. Oh, it's the same thing. Is it the same thing? All right. So providing all of that in one, let's see. Let's see what these are going to look like. I'm also going to see if I can create in Manus. I was able to create a website, like a shareable website with a URL and everything. I'm going to see if I can do the same. Actually, I should be able to do the same in Gemini. Just make this into a one-pager website. Now we can add into like our, use as our portfolio, so to speak. I will say when it comes down to. You know, I said, hey, I want to do consulting style images. I think McKinsey, BCG. It was very good. Like we saw the before and afters. It was very good at the initial ones. It was, you know, very small changes that I wanted to do. You know, I've had like ask ChatGPT and stuff to like, hey, make this consulting style and stuff like that before and not very good. Or it took a while or it took like so many prompts. Those are, that was a while ago, but. So far, this, you know, Spark is very good so far. Just wish you can talk to or interact with it while it's in mid-prompt. That is what you can do with Manus. Okay. So now what I'm going to do is like, okay, can you build a, you know, one-pager website for one-pager portfolio to display our Work. All right. Thinking it through. Let's see. Can you build? I'm just going to ask it to build a one pager. See where it goes. Also, you can have it do something and make it into even like a PDF. I've seen people do some really nice PDFs for their portfolio. Excellent. Like there's so many ways to make a portfolio, you guys. That's like not hard. A PDF is probably more preferred because most likely if it's a website, you have to like pay extra. For like hosting capabilities and like we were talking about just for the sake of a portfolio. I don't think it's worth it if you want to or not. No, come on. That's not what I want. Like, no, one-pager means everything is on the page. Uh, no links to have to view the images. Thinking it through, thinking it through, and then we'll see if it can also create a website. So we'll have multiple options, a PDF and then a website. That's good. See, no, it's not doing it right. It's still, oh my gosh. No, hiring manager would still have to click the links to see the images. They need to be displayed. On the doc, you know, exported as a PDF. Hopefully it gets it right this time. All right. Now I understand the images must be rendered on the document for PDF, a limitation I'm addressing now. All right, cool. All right. What I will do is I will have the link to the PDF publicly in the description for those who want to look at it as a finished product. And I will see if it like probably use Canva or Canva to create like a website. One pager, and so we'll have two different outputs. Hopefully. All right. This looks ugly. Actually, I don't know why it cut off. What was this? But even with this so far, like you could even just take this and just paste in what we have. This I liked having on the doc though. This visual. What I might have tell it to do is just like export in Google Docs first. Dude, where are the images? Add the dang images. Insert the images. Why is it like? Why is it so hard for it to insert the images on the thing? Like, I, I'm like shocked. It's like, what's going on? Is this a limitation of Spark? I know it's not. Is this a limitation of Spark? I know I've, I've made stuff like this. I've done stuff like this in Manus where it would output stuff in Word doc and it would have the images on there. You know, if you wanted to like paste in whatever you can, we're just going to have, see the limitations and things like that. We're just exploring. Um, let's try this again. Oh my gosh. Open with Google Docs. Can I add this up here? All right. Okay, and then what I would do is like, you can just like make these numbers bold. Now what I would do is probably like remove the 74, 25. All right, well, let's see if we can use it to make— I think it's called Canvas on Gemini because it can make its own like website with the URL. So here's one. We're going to save this as a PDF. See what this looks like. It's still— I don't know why this one is like— why is this so weird? And then what I would do is also I'm just going to have it remove this line. Okay. It's an HTML. All right. I guess that's a limitation here. I know in Gemini there was like a Canvas thing. Never mind. Let me go back up here and say, for the image on finding 2, the header and subheadline are overlapping and you cannot read the subheadline. Also remove benchmark line. And again, in real world, you can just talk to a user of Microchrome. Just typing through it for the sake of the video, but You can definitely use the microphone. It's definitely easier. Okay. I like the way, like the layout of this. I'm going to say I liked this layout the best. Please just render the images. So I like the layout of this the best in terms of like even exporting it as a PDF. Oh, I want the data quality ledger on there too. I mean, now this is also like preference, like in terms of like how each are like, you can have like a one page, like a Like this type of, like, this is fine. Or you can do it kind of this way. Like if we had a visual, the metric, et cetera, either one is fine. It's just going to be preference. I think for this, for the sake of like what's going to be our final thing. Uh, let's see. What I also like actually about this is like the, it's actually, it just seems more simple. To be honest, as well. Like, it doesn't look too over-engineered. When I say too over-engineered, it's just like, it doesn't look too overly polished. Even like when it comes down to like your portfolio, like this is a lot of text. This is my favorite way. Like, this is good actually. For number 3, you know, for finding 1, for finding 1, I'll say remove, but like the labels, the actual value labels, the value labels are not Easily read, move them on the outside. For this one, again, you can't see. So I'm just going to tell it to just move the labels on the outside. But again, either way, we have everything. Now, one thing I will add, I mentioned this earlier, I want to do a visual. I want it to make a visual bottleneck, like visual representation of the bottleneck and also a visual representation. Representation of the schema of the tables. So those are two things I want to make sure that we add, and then then we'll be good to go. But now add in a visual representation of the bottleneck in the process, and also a visual of the schema. Ah, what's happening? I reached my limit. Okay, well, I've reached my limit. That's, that's one thing. Let's see if this is the most recent one, and then we're just gonna publish it. And then I'm not gonna have this workspace. I don't, you don't need to see all that. Executive summary. Good. I like this. This is here on time to deliver. Okay, so this, okay, so I'm gonna export this as a PDF, and this will be below for people to look at for review. So yeah, that's an end-to-end in doing a project with Google's new AI Spark. We saw the, you know, the good stuff, the bad stuff. Overall, like I said, if there's anything using AI, especially the agents to find and source high-value projects for your portfolio, absolutely number one. You should definitely do that. You don't— no one needs to use Kaggle anymore. That's just Number one, and you can design it and structure it for your target domain. And again, we learned some really good. I mean, there's some really really good domain expertise and like learnings that were we've done just so far. Just this again small sample size of a project. Ideally, you can spend like you know two days on certain things. Something like this. But good bottleneck analysis. Now, if you want more videos like this. Please let me know. I can make this more of a, a series, you know, pretty much AI agents for data analysts. If you ever wanna work with me and get one-on-one coaching to launch your data analytics career, click the link in the description. Would love to work with you, and I'll see you in the next one.