How I Use Claude Code as a Data Analyst (10 Real Use Cases)

0:00Data analyst role is changing. The

0:02analyst pulling ahead aren't the ones

0:04spending more time at learning a sequel

0:05or Python. They are the ones that are

0:07using AI to do in 5 minutes what used to

0:10take a 5 hours. I've been in this space

0:13for years and I've never seen a tool

0:14change how much I actually work as fast

0:16as Claude Code did. And in this video,

0:19I'll show you exactly why. 10 real

0:21workflows from cleaning up messy data to

0:23building internal tools. And by the very

0:26end, you'll see why this is the skill

0:27gap that is opening up right now in the

0:29data space. So, we're going to start off

0:32with probably the easiest use case and

0:34this is going to be data clean up. Every

0:36single analyst is going to have a

0:37version of this file. You get a CSV from

0:40someone in your company that's data

0:41illiterate and it is just a complete

0:45mess. There's tons of issues with it and

0:47you're like, why did you send me this

0:50file? You couldn't clean it up? Uh for

0:52example, if you look at order date, we

0:54have multiple date formats. 01, okay,

0:56cool. Then we start 2024, then we start

0:58with January. Like, what are we doing

1:00over here? Then we go into something

1:02like region, New York capitalized, not

1:05capitalized. Then we have a blank region

1:06over here. Then we have different things

1:09over here like a a repeated customer or

1:11duplicated rows or missing data. It's

1:14This is just not good. In fact, there's

1:16seven distinct issues

1:18with this particular file that we're

1:20going to want to to

1:21clean up. So, I want to send this over

1:24to Claude Code and see if it can solve

1:26the issue. Typically real scenario,

1:28either you're going to go and clean this

1:30up manually in Google Sheets or Excel or

1:32you're going to write some Python code

1:34to go out there and really fix up this

1:36file. You're going to go back and forth

1:38and hey, do we actually need this row or

1:40not?

1:41And even before we do any analysis, it

1:44turns out to be 1, 2, 3 hours. Now,

1:46obviously this file over here super

1:48simplistic. We're looking at 30 rows.

1:50Real scenario, you're going to be

1:51working with something that has hundreds

1:53if not thousands or tens of thousands or

1:55hundreds of thousands of data points,

1:57but I want to show you how powerful

2:00Claude is and all I'm going to do is

2:02just describe in plain English what I

2:04particularly want it to do. So,

2:07I'm going to close out of Google Sheets

2:09over here and you don't need to load

2:11this spreadsheet up in Sheets if you

2:13want. Uh you could actually just go in

2:15load this in Excel as well. Does not

2:17matter.

2:18So, I'm going to close out of Google

2:19Sheets and then I'll load up Claude

2:21Code. Now, the version of Claude Code

2:23I'm using is directly in VS Code. Uh for

2:26each of their demo examples, I'm just

2:28going to load up the folder with our

2:30data points that we particularly want to

2:32take a look at or whatever file

2:34and I'm going to use Claude Code over

2:36here on the side. So, this isn't going

2:39to be a tutorial on how to set up Claude

2:41Code in VS Code. If you guys want to see

2:44that, I'm happy to make it that

2:46tutorial. Uh just the assumption is you

2:48have Claude Code readily available to

2:51go. All right. So, what I'm going to go

2:53over here is say, "I have a messy sales

2:57CSV that I need to clean up and analyze.

2:59The file is messy sales data.csv", which

3:02you guys can see that is over here.

3:05And obviously, if you're if you're

3:07looking at other stuff, yes, I already

3:08have the prompt written and I already

3:10have my notes for the section.

3:12And what I'm going to say is, "Please

3:13load it and tell me everything that's

3:15wrong with it. Fix all the issues you

3:16find, save a clean version, and give me

3:19a summary of total revenue by region by

3:21sales rep." So, let's grab this prompt.

3:24I'm just going to throw it over here.

3:28And uh

3:30you can see

3:32it's the exact thing I've already said.

3:35I'm going to make sure that ask before

3:36any edits

3:39as my specific mode.

3:41And then I'm just going to run this over

3:42here. So, just click on this.

3:45I should probably move myself out of the

3:46way so you guys can see that and uh

3:49we'll run it.

3:50And I'm not going to narrate over every

3:52single time that it goes through, but

3:53you can see that it's thinking, it's

3:55going through. It also found seven

3:58issues as I addressed, and it says allow

4:01right over here. That's fine. I'm going

4:03to say yes,

4:05and it looks like it's done. It took me

4:07probably about I'd say a minute or two.

4:12So, what you can see, I have my initial

4:14prompt over here.

4:16Then it's reading in the file. Then it

4:18says, "Here's what I found. Let me write

4:19a Python to fix everything." So, the

4:21issues that it found, number one,

4:23heading white space, customer name

4:25trailing space, order date leading

4:26space, which obviously was hard to see

4:28in Google Sheets, but uh it found that

4:30over here.

4:31Inconsistent casings on names, region,

4:33sales, randomly all caps.

4:36Three different date formats.

4:39Revenue as currency.

4:42Missing values.

4:44Confirmed duplicate rows and region

4:47inconsistencies.

4:49So, then it went through over here.

4:51It started cleaning up our issues. So,

4:54you can take a look at all the

4:55individual steps.

4:58Everything looks clean. Here's the full

4:59rundown.

5:01You can see

5:02revenue as currency strings. We have our

5:04revenue summary as well, which, you

5:06know, we asked it to clean up our data.

5:08So, load it in, save all the fix all the

5:10issues you find, save a clean version,

5:12and then give me a summary. You can see

5:14we have a clean version over here, which

5:16I'm going to load up in Google Sheets

5:17here for a second with you guys.

5:20But then we have the revenue by region.

5:21So, we have Southeast, Midwest, West

5:24Coast, Northeast, New York, and then by

5:27sales rep, each of the sales reps as

5:29well over here. And then it says, "Note,

5:31row 1027 has no sales rep assigned, and

5:34rows 1009 and 1021 have missing revenue.

5:36Those amounts are excluded from the

5:38totals above."

5:39I do love that note as well. I mean,

5:41that's something you would want to give

5:43to a specific stakeholder. So, let Let

5:45load this up really quick in Sheets, and

5:47uh

5:48All right, so now we have our clean one

5:50over here. Date format

5:52all consistent across the board.

5:54Revenue, it's stripped away the dollar

5:56amounts, which maybe you want to have

5:57that in over here, but this was casted

5:59as a string, uh which makes things

6:01tough. Obviously, it did automatically

6:03convert this when I uploaded it into

6:05Google Sheets, but as it's CSV and

6:07normally it wasn't.

6:09And then you can look at capitalization,

6:10like New York, this looks good. Uh

6:12customer name, this looks good as well.

6:14Uh removed our duplicate row, and uh we

6:17still have the missing revenues, which

6:19it called out. What's really nice,

6:20obviously, is it just summed up all

6:22these different regions and as sales

6:23reps. So, you know, honestly, it

6:26wouldn't really take that long to write

6:28a line of Python Pandas to get that

6:30information or just to do some filters.

6:32You could go over here and date filter,

6:34you could go, "Hey, I'm going to create

6:35a filter. Let's look at a region of New

6:37York." So, we'll say clear all, New

6:39York. And some people would be like,

6:41"Okay, well, New York has 5,470."

6:44Well, you go back over here,

6:47already done for you. So, you don't even

6:48have to worry about that side of it. So,

6:50like, obviously, this is our first

6:52example of 10 that we're going to go

6:54through today, but already, I hope that

6:57you are seeing how valuable it is. All

7:00right, let's jump into example number

7:03two. So, example number two is just

7:04going to expand upon what we did a

7:06little bit earlier. This time, we're

7:08going to take a spreadsheet, and we want

7:10to find data from it. And

7:13this is something I think every analyst

7:15still do on a weekly basis, if not a

7:17daily basis. And there's a lot of

7:19different approaches, you know, you can

7:20go over and write Excel formulas, say

7:22equal sign and then filter for specific

7:25criteria. You could go the old school

7:27route,

7:28do a filter, and then just go over here

7:30and say, "Hey, I just want to find

7:31what's going on with software licenses.

7:33You know, how many units did we sell?"

7:35Highlight it. Okay, we sold 2390. Or if

7:38you're a little bit more advanced, you

7:40can use something like a Python Pandas.

7:42And personally, I've used Python Pandas

7:44over the last few years and it's saved a

7:46ton of time. But maybe Python is

7:49intimidating for you and you want to see

7:52the efficiencies. Well, Cloud Code could

7:54do all of that for you without even

7:56knowing how to write Python. Now, what I

7:59will say is you should eventually try to

8:01learn the basics of Python Pandas. It

8:02will help you out, but we can automate

8:05this. So, we're going to actually go

8:06through three different prompts with

8:09this data set. Uh we're going to do some

8:11basic aggregations. We're going to do a

8:13pivot table and then we're going to do a

8:15performance analysis. So, I'm going to

8:17go into VS Code. You can see I already

8:19have this loaded in.

8:22And what we're going to do is ask our

8:24first prompt. So,

8:26I'm going to

8:27myself from here and let's write this

8:30out.

8:31So, essentially what I'm going to say

8:35I have monthly sales CSV and you can see

8:38that's a CSV file. Group by region and

8:40show me total revenue and average deal

8:42size for each.

8:45And we're going to run that.

8:48I'll be back when it's ready. All right,

8:50it is back with these results. Probably

8:53took 2 minutes, give or take. And

8:56obviously you can see it's thinking. It

8:58read our file and you'll see it actually

9:01brought in Pandas. So, pip install

9:03Pandas and here are the results. So, we

9:06have region over here, Midwest, total

9:08revenue as well as the average deal

9:10size, Northeast, Southeast, and West

9:13Coast. And I didn't even ask for it, but

9:15it gave us key takeaways. So, it says

9:17Southeast leads in total revenue at 1.7

9:20million. West Coast has the highest

9:21average deal size, nearly 2.4 the

9:23Midwest average. Midwest lacks

9:26significantly in both metrics with

9:28608,000 less revenue than the next

9:31lowest region. The code used was a

9:33simple group by and agg with the named

9:35aggregations, the Pandas equivalent of a

9:37pivot table summary.

9:40So,

9:41obviously this wouldn't take super long

9:43to write out the Python code for it,

9:44which is why I always say to learn

9:46Python Pandas as probably early as

9:48possible within your data career.

9:50However, it was able to solve this

9:52relatively fast. Okay. So, now what I'm

9:54going to ask it to do is build out a

9:56pivot table. Again, you could solve

9:59pivot table pretty easily within Pandas

10:01or if you're using Excel or Google

10:02Sheets, but I'm going to say now create

10:04a pivot table showing each sales rep as

10:06rows and each product as columns with

10:08total revenue as values, just like I

10:10would do in Excel. So, that is the exact

10:12prompt. I'll send that over here.

10:16And uh

10:17yeah. All right, this was actually even

10:19faster.

10:20And you'll see it did all its thinking

10:22over here. It showed us exactly what it

10:24was thinking. And what we have is our

10:27pivot table. So, we asked over here,

10:30showing each sales rep as rows and each

10:31product as columns.

10:34So, we have

10:36Angela, David, James, Jennifer,

10:40Marcus,

10:42Patricia, Robert, Sarah, and our grand

10:45totals at the bottom. So, 472 across the

10:47board. And this is for consulting, then

10:50we have hardware. And then if I expand

10:52this out, this is actually way nicer.

10:55Uh I don't know why I had this so small

10:56for you guys. We're going to make this

10:58essentially full screen.

11:01Over there.

11:02And uh you can see we have hardware,

11:04software licenses, and grand total. The

11:06key function, pd.pivot_table, maps

11:08directly to Excel insert pivot table.

11:10And then it says rows index equals sales

11:12rep, columns is product values

11:14summarize. So, a few things to notice in

11:16the data, reps specialize, Angela and

11:19David are pure software license sellers,

11:21while James Wilson is hardware only. And

11:23let's see if that's the case. We go to

11:25James Wilson, and you can see hardware

11:27with no consulting, no software. And

11:30it says Angela is also the same. So,

11:33Angela is software only.

11:36Yep, David software only.

11:38So, awesome insights across the board

11:42over there. Let's actually give it a

11:43third request now, and we're going to

11:46ask what reps are below their target.

11:48So, if you go back to our spreadsheet,

11:50you can see that we have a target over

11:52here as well as the total revenue. So,

11:55you know, it's it's a very

11:57blank ask, but we're going to see what

12:00definition it actually gives us. So,

12:03I want to ask it what reps are below the

12:06target, show me that gap. And we're

12:09going to go over here and run that as

12:11well. Then we have the information. Only

12:13one sales rep is below target. By the

12:15way, this area that took like 10

12:16seconds. Uh Marcus Johnson, it says

12:18revenue 822, target 960, minus 137.

12:22Marcus only rep missing his numbers,

12:24137k short at 85%. Analysis above or

12:27below, want to see the full leaderboard

12:29with all reps. And then I should ask is

12:31this by

12:33by month and region as well, because I'm

12:36sure some of these have missed, and

12:37let's just do a quick sanity check. And

12:40you can see right over here that was

12:41Marcus in January 2024.

12:45Obviously, he did get called out across

12:47the board, but we're not seeing the

12:49individual rows.

12:51So, I just prompted it again, and

12:53sometimes it won't get everything

12:54correct. I I mean, I think it did a good

12:55job here, but I just said you didn't

12:57give me the individual rows. Also, I

12:59want to see individual months a rep

13:00missed. What if they just had one good

13:02month mask a ton of bad months? And

13:03that's possible. Let's say someone was

13:05100,000 over target in one month, um but

13:08they had four other months they were

13:09down 20,000 each. Well, in net wise,

13:12they'd be plus 20%, but they only hit

13:14target one specific month. Um so, I'm

13:17just asking if it can do some analysis

13:19based around it. So, I want to see those

13:20individual rows, and then analysis based

13:22around it. So,

13:24I'll send that over. Let's see how long

13:25it takes, and uh gave these results.

13:28Okay, so

13:30here's the full picture and we can see

13:32Marcus missed, you know, four in a row.

13:35Then we have Jennifer that missed two

13:37and then we have Patricia and we have

13:39Robert. And you can see the gap across

13:42the board. So this was zero flat. I

13:44don't know if we probably should have

13:45had zero flat as a miss. I think that's

13:48a hallucination.

13:50Um but the other ones are accurate. This

13:53is summary scorecard. Months hit one,

13:56months missed three, hit rate. So

13:58actually did get that correct. I don't

13:59know why we got to see this one over

14:01here. Just a very small bug.

14:04Obviously double check the work that is

14:06being done.

14:07Jennifer Lee two two, Patricia three

14:09one. So you can see hit rate, overall

14:11gap, and also the worst monthly. Your

14:13instinct was right to dig deeper. That

14:15grid view hides a lot.

14:17Obviously as a data analyst you should

14:18know to always poke into the data. Uh

14:21things like this should be pretty

14:23apparent.

14:24It says Marcus Johnson is in free fall.

14:26He only broke even January then missed

14:27every month after that with an

14:28accelerating gap. Jennifer Lee

14:31overall positive but each missed two

14:32four months. They're inconsistent, not

14:34safety. Both had one big miss but

14:36otherwise are solid contributors.

14:39So hopefully

14:40showed you another way to really start

14:42diving into the data. Honestly without

14:45Claude this would have taken me a little

14:46bit longer to build even something out

14:48like this. I would have had to think

14:50about it, you know, what's the hit rate

14:52across the board and do a little bit of

14:54Python code.

14:56Definitely doable but but obviously

14:58we're talking about minutes versus

14:59hours. Because it even gave us a

15:00write-up which is really nice. And

15:03obviously there's a lot that you could

15:04do with this too. You could eventually

15:05turn this into presentations or

15:07documents. Talk about some of that a

15:09little bit later on uh within this

15:11particular video. But I think we are

15:12ready to move into our next example.

15:14Here's something that's happened to

15:15every analyst. You're about to present a

15:16report to your manager. You hit send.

15:19Five minutes into your meeting, they

15:21spot an error, a number that just

15:23doesn't make sense, a missing value or a

15:25duplicate entry, and suddenly you're

15:28scrambling. So, what if there's a way to

15:30catch any of that type of stuff before

15:32the meeting? Not just visually scanning,

15:34but actually analyzing the data for any

15:36sort of anomalies, missing values,

15:38inconsistencies. That's what we're going

15:40to do in this section. Essentially, what

15:42we're going to have is Claude Code as

15:44our second set of eyes. Within this CSV

15:47file, there's going to be six separate

15:49errors that I wanted to catch. Do want

15:51to note that this isn't just focused on

15:54CSVs. Within the Claude ecosystem, you

15:56can also have it take a look at

15:57documents as well as PowerPoint

15:59presentations. Just for ease of use

16:02within this particular video, I'm going

16:04to use a CSV file. So, let me show you

16:06what the CSV looks like.

16:08So, we have our date, we have our

16:10customer, customer name, region,

16:12product, revenue, units, sales rep, as

16:15well as margin. And uh

16:18I'll send you over that quote here in a

16:20second that we're going to send uh to

16:22Claude Code, but

16:24we have a huge outlier in this data set.

16:26Take a look at this revenue. Like, that

16:29just is giant. And that's one of the

16:32things that we want to see. Uh we also

16:34have a negative revenue value.

16:36And we have some missing values across

16:39the board.

16:40There's a duplicate customer ID, which

16:42you'll see that here in a minute.

16:44There's inconsistencies about margin

16:46percentages with regions. Uh we have a

16:48wrong quarter as well. So,

16:51I want us to catch all six of these

16:53errors. And uh

16:56Yeah, so what we have is Claude Code.

16:58I'm going to

17:00type in this prompt, and then I'll just

17:02move my uh screen away. So, that way I

17:05am not in the way.

17:06So, our prompt we have

17:09I have a Q1 revenue report that I need

17:12to present to my manager in an hour.

17:15Before I do, can you check this data for

17:17anything suspicious or wrong flag

17:19outliers, missing values, duplicates,

17:20anything that doesn't look right?

17:22Obviously, we have that spreadsheet over

17:24here called Q1 revenue report. I'm going

17:26to click send and we'll be back when

17:29it's ready.

17:32All right, it's ready. So,

17:34critical issues. Number one, it spotted

17:36that revenue outlier, $980,000.

17:39It also spotted this negative revenue.

17:42And then it spotted also a wrong date.

17:46Year-end corp date.

17:472024 12 15. December is not Q1 by any

17:50fiscal calendar.

17:52And that is the one that we wanted to

17:54take a look at.

17:56You can see

17:58have that right there. Should not be

18:00present in this data set. I guess it's

18:02kind of easy to then spot, but I'm glad

18:05that was caught.

18:07We have missing revenue,

18:09missing margin percentage over here. And

18:11then it says minor observation April

18:13date included. This data set runs

18:15through 2024 4 28. Your Q1 is January to

18:18March. Rows 33 through 43 shouldn't be

18:21here. Fiscal Q1 runs through April, it's

18:23fine. Uh three critical issues, 980K

18:26outlier, negative revenue, and December

18:28date. So, going back to my notes, 980K

18:31was caught.

18:33The negative revenue was caught. Missing

18:35values.

18:36Uh it talked about row 12. And there I

18:41believe is a few other ones with margin

18:42percentage. Let me just double check.

18:45Well, we have that over here, the

18:46missing margin percentage, so we're fine

18:48there.

18:49It did not catch anything about

18:51duplicates, though. Um there are some

18:54duplicate issues. So, I should ask on

18:57that, "Didn't you miss?" And by the way,

19:01this is me already like cherry-picking

19:02the data. I already know that there is

19:04duplicate issues.

19:05But, one of the things you'll notice the

19:07more that you use Claude is always ask

19:10it a second time or a time. Like it will

19:12not get everything correct the first

19:14time, which I think is why some people

19:16absolutely do not like AI. They're like,

19:18"It's terrible." Well, it's not terrible

19:20by any means, but you need to constantly

19:22prompt it to get the best results. So,

19:25I'm going to say, "You missed

19:29the duplicate values."

19:35Were there any other

19:41inconsistencies?

19:45My terrible spelling over there.

19:54Um we have a few other inconsistent

19:56areas over there, and then we already

19:58talked about wrong quarter dates.

20:00Cool. So, I mean, it missed duplicate

20:03values. I'm just going to ask this over

20:05here.

20:08Regardless, five of the six main issues

20:10were caught. That's pretty good.

20:12And obviously that would save us a

20:14decent amount of time.

20:15You know, even with like my code that I

20:17write for full production,

20:19I will send this into Claude just to do

20:21a double-check at the very end. Uh to

20:23tidy things up. Sometimes, you know, if

20:25I'm writing pandas code, it might not be

20:27the most efficient, so I'll have it

20:30check through that, see if I have any

20:31specific errors, and uh create

20:34something. But, we'll talk about the

20:35coding review side of it a little bit

20:37later also in this particular video. So,

20:40the pushback, it definitely took a lot

20:41longer uh this time across the board,

20:44but

20:45uh what I just want to show you,

20:47it says on duplicates what's actually

20:49there. So,

20:50talks about some duplicates over here,

20:52additional inconsistencies I should have

20:54flagged.

20:55So, that's 20% drop over 3 months, the

20:57same customer and product with no

20:59obvious volume justification. It could

21:00be an undocumented discount or pricing

21:02error. Love to see that. Row 11 unit

21:05count is also not higher, not just the

21:06revenue.

21:07Our 20 to 50 units, row 11 has 500

21:10units, 10x the next highest. Given Drake

21:12only covers one account. He appears

21:14exclusively over here.

21:16So, yeah.

21:17Um pretty good job across the board

21:20taking a look at this data set. Let's

21:22now jump into our next example. Here's

21:25what every analyst knows, your data just

21:27doesn't live in one place. The CRM has

21:29customer contracts, billing has

21:31invoices, marketing has engagement

21:33metrics, finances is something and your

21:36boss might ask you for a customer health

21:39data set. And this has to combine all of

21:41these sources together at once. You can

21:44do this in Excel and it can be a mess.

21:46You're going to have to use a VLOOKUP,

21:47index matches, as well as a manual fuzzy

21:50matching because names aren't going to

21:51line up and it could take you hours.

21:53Even if you go through the Python route,

21:55you have to make sure things are going

21:56to line up one-to-one on your joins and

21:59you're going to still spend a decent

22:00amount of time cleaning up your data.

22:02So, I want to show you how we could do

22:04this within Applied AI Code. We're going

22:06to have three real files, CRM data,

22:08billing data, as well as marketing

22:10engagement. They all use different keys.

22:12The names aren't going to match exactly

22:14and we're going to combine them all

22:16together in just one request. So, let's

22:19jump into the files themselves.

22:21First up, we have our billing data.

22:24Invoice date,

22:25sorry, invoice ID, customer ID, invoice

22:28date, amount, status, as well as our

22:30payment date. Then we have our CRM

22:32customers. So, we have 30 different

22:34customers over here.

22:36Name, industry, account owner, start

22:38date, contract date, as well as their

22:40customer tier. And then we have

22:42marketing metrics, company name, last

22:44email open, if they attended a webinar,

22:46an NPS score, and then also last

22:48contacted. So,

22:51let's jump into code and run this

22:53through. So, I have three files, CRM

22:55data, billing data, and marketing

22:57engagement. I need to combine them into

22:58one view showing each customer's

23:00contract value, whether they have

23:01overdue invoices, and their NPS score.

23:03The marketing file uses the company name

23:05instead of customer ID, and the names

23:07aren't exactly the same. So, obviously

23:10that type of context really helps if

23:11things aren't going to be lined up

23:13one-to-one. If they were lined up

23:14one-to-one, obviously you could just

23:16write some Python code and solve that

23:18pretty easily, but this is asking us to

23:20clean up the names as well as knowing

23:22some a little bit additional context in

23:24each of these files. In a real scenario,

23:26I'd probably make this prompt a little

23:28bit longer, a little bit more thorough,

23:30but since this is just a demo for you

23:31guys, let's run it. Okay, so it took

23:33literally about a minute or so, and

23:37now it gave us the results of overdue

23:39invoices. So, you can see the customer,

23:40the contract value, the overdue amount,

23:42as well as the NPS score associated for

23:45the customers, but it didn't give me uh

23:48a spreadsheet. We do have this over

23:49here, the script save combined data.py,

23:51which is Python standard library, no

23:53external dependencies. Um I'm just going

23:56to ask it, "Can you build me a CSV

23:58file?"

23:59I guess I did tell it that I just want

24:00overdue stuff, so maybe I should just

24:02add it in general, but you know, it

24:04would be good to see the contract value

24:06and zero dollar amounts. And obviously

24:09we could just filter that out if we

24:10really wanted to. All right, so we have

24:12something over here called customer

24:13combined view.csv, and I'm just going to

24:15load this up in spreadsheet now, and

24:17we'll take a look.

24:19All right, so this is our spreadsheet.

24:21We'll put the camera back on over here.

24:23We have the customer ID, name, industry,

24:26account owner, tier, contract value, has

24:29overdue invoices with the yes or no,

24:30which we could just easily filter. We

24:32could just go data

24:34create a filter,

24:35and then we'll just say yes, and you can

24:37see we have these six over here, overdue

24:39amount. We didn't have the NPS score or

24:41webinar attended for the last one, and

24:43then we had the last email, as well as

24:45last contact date. Literally just

24:47minutes instead of hours for this build.

24:50So,

24:51now let's jump into our next example.

24:53We've got messy data with phone numbers

24:55in five different formats. You know,

24:57regex could literally solve this in 2

25:00minutes. We also know that regex is

25:02annoying to write and you'll at least

25:04spend 30 minutes to an hour debugging

25:07and Googling, "Hey, should I include

25:08that parentheses or not?" And you just

25:11don't do it. Instead, you go through

25:13this spreadsheet, you find and replace,

25:15and do it manually. Today, we're going

25:18to be fixing that with Claude Code.

25:20We're going to go actually go through

25:21three different data points we're going

25:23to clean up. Let me show you the

25:25spreadsheet and then we'll jump right

25:26in.

25:27Okay. So, what we have over here is a

25:30full name column, which honestly doesn't

25:31really matter that much. Then we have

25:33this phone number. This is ideally what

25:35we're going to clean up because these

25:37phone numbers, they're all over the

25:39place.

25:40We also have different company names.

25:41You can see like LLC versus LLC, Inc. uh

25:45for incorporated. We're going to clean

25:47that up. And then lastly, within our

25:49notes over here, we sometimes have the

25:51budget for these companies, but

25:54it's in the text and to extract that

25:57out. Those are our three goals. We're

25:59going to actually write a prompt to do

26:00that. And uh let me jump into code.

26:03Okay.

26:05Away goes the webcam. We're going to

26:06scroll this over here.

26:09Now, I've already written this prompt.

26:10So,

26:12I'm just going to tweak it maybe a

26:14little bit, but I'll read this out for

26:16you guys.

26:17And maybe we'll expand this out. So, I

26:19have a CSV file with messy contact data.

26:23And let me actually grab the exact name.

26:27It is raw contacts.

26:30Let me throw that over here.

26:32Raw.

26:33Should be able to pick it up. That's

26:34fine. The phone number column has

26:36different formats, so I just put some of

26:37the formats over here. I need them all

26:39standardized to this. So, specify the

26:41exact format.

26:42By the notes column in raw contacts CSV,

26:45notes column contains revenue

26:47information in different formats.

26:48Extract all the dollar amounts and

26:49create a new column called estimated

26:51deal value with just numeric values.

26:53Multiple amounts found, take the largest

26:54one. If no amount found, use null. The

26:56company column has inconsistent

26:58formatting. I created a script that

27:01removes all punctuation, title cases,

27:05and removes the trailing punctuation

27:07extra spaces.

27:09Okay.

27:10I see that over here.

27:13Let's see. Update

27:16the column

27:17with the clean information.

27:22And then lastly, I say once

27:24these are fixed, save it as clean raw

27:26contacts.csv.

27:28I should say as a new

27:31file called So, you get the point. Like

27:33I

27:34give it examples. Like this is the bad

27:36data. This is how I want to see the

27:38format.

27:39This is the type of data we'll see in

27:41this column. What do I want to extract

27:42out? Edge cases as well. And lastly,

27:45again, more examples of what's

27:46happening. You know, remove all

27:48punctuation, title case, remove trailing

27:50punctuation.

27:52Gives that example. And I just want to

27:53create a new spreadsheet. So,

27:55uh very painstakingly

27:58I had to write out regex on here.

28:00And I'd rather just clean it. Okay. So,

28:02run it. And once it's done, we'll be

28:04back. Okay. So, it took about 10 minutes

28:06and we're back. So, first our phone

28:09numbers, well, these have standardized

28:11format. So, I think that looks pretty

28:13good. Uh the next we were taking a look

28:15at the company. You can see LLC has been

28:17fixed.

28:20So, that's been pretty good.

28:22We fixed the spacing on stuff.

28:25Now, this is all lower case.

28:27You can see global solutions.

28:31LLC is combined.

28:35So, this looks a lot better as well.

28:38And lastly, estimated deal value, you

28:41know, 45K, we got that. 120,000, 120.

28:45And we were able to extract all this.

28:48As it was going, obviously it was giving

28:50us Python code

28:52directly over here.

28:53And you can see we actually have our

28:55Python code saved.

28:57So you can see pandas and import RE,

29:00regular expression.

29:02And this one turned out to be a lot more

29:05code than I initially thought it was

29:06going to be, but that's okay. About 140

29:09lines that was developed, but uh

29:11we have this spreadsheet again. We need

29:13to clean it up anytime we just can run

29:15that code. So

29:17pretty awesome.

29:18I hate writing regex. I think regex is

29:21super powerful. Um but ever since I've

29:23been using just AI in general, I have

29:26not written regex by hand. I just send

29:28it over to an LLM, say what I want to

29:30extract, and it saves a ton of time. So

29:36let's jump into our next section. SQL

29:38can have a special kind of frustration.

29:40Your query runs perfectly, there's zero

29:43syntax errors, no exceptions, but it

29:45gives you the wrong answers. You've got

29:47a revenue report that says you made

29:48three times more than you actually did,

29:50your customer ranking is grouping

29:52everyone together instead of by tier, or

29:54your month-over-month growth is just

29:56jumping around randomly. Your query runs

29:59though, it's not broken, it just gives

30:01you the wrong answer. And sometimes

30:04debugging that is not really fun. So in

30:07this section we're going to cover is the

30:09first part of this, which is debugging

30:11these queries to finding and

30:12understanding specific logic errors

30:14within SQL. And the second part is going

30:17to be optimization, because even if you

30:19write a query that runs, doesn't 100%

30:22mean it is fully optimized. It might

30:24take 5, 10 minutes to run, and reality

30:27if you work that query over again, you

30:29might be able to get it down under 30

30:31seconds. So in addition to just fixing

30:34our code, I'm also going to have a

30:35Claude explain what happened in our

30:38specific queries, so that way we can

30:40turn this frustration into a learning

30:42moment. Okay. So, we have three files

30:45for this section. First, we have our

30:47schema SQL. So, this is an e-commerce

30:50database schema

30:51looking at PostgreSQL. Obviously,

30:53there's different flavors of SQL out

30:55there, but this is what we use for this

30:56video. So, we have a customers table,

30:59products table, sales reps, orders, and

31:02then order items.

31:04Then we have sample data for each of

31:06these over here.

31:08Then we have our broken queries. So, we

31:11have multiple broken queries all

31:12together. Maybe you will only want run

31:15one query at a time in a real scenario,

31:18but I just want to show you that it can

31:19handle multiple queries. So, maybe you

31:21put in multiple at a time either way.

31:23So, we have our first one over here.

31:26Our second one, as well as our third

31:29one. In addition, we have a second

31:30section which is a slow query. So, this

31:33query works, but takes 4 minutes to run.

31:35I need it faster.

31:36And

31:37take a look at that over there. Okay.

31:39So, let me close these out.

31:44And we're going to start off with our

31:46first prompt, which is going to be

31:47fixing the queries. So, let me copy over

31:52our prompt, and then I will read through

31:54that. I'm also just going to get out of

31:55the way.

31:57So, that way we can fully see this.

31:59Okay.

32:00So,

32:02I have three SQL queries that are giving

32:04me wrong results. The schema is in

32:05schema SQL, and I have sample data

32:07loaded. Here are the queries in the

32:09broken queries, and I talk a little bit

32:11about each of the queries. For each one,

32:13identify the bug, ex- explain Oops.

32:17Explain why it's producing the wrong

32:18results, provide the corrected query. If

32:20you can, show what buggy output was

32:22versus what it should be. Getting your

32:23head around is why each bug happened,

32:25not just getting it to run correctly. I

32:27think a lot of people go out there just

32:28slam SQL code or Python code into Cloud,

32:31get the right answer, and they never

32:32learn anything. So, it's really

32:34important that you understand where your

32:35mistakes are.

32:37And we'll run this. All right, and

32:39honestly this was one of the fastest

32:40ones today, probably like

32:4245 seconds to solve these.

32:45Now obviously these are very basic SQL

32:46queries in comparison to what happens in

32:48the real world and in real scenarios

32:51sometimes you're going to have to go

32:52back and forth and prompt this in

32:53multiple times. Again, this is just a

32:55video. So, query one revenue by region

32:58fan out from unnecessary join. The bug

33:00the query joins orders order items but

33:03order total amount already stores the

33:04full order. Each order gets duplicated

33:06once per item.

33:08Why this happens is it tells us why this

33:10exactly happens.

33:12Buggy versus correct results.

33:15And you can see our corrected query. It

33:17says drop the join entirely, it was

33:18never needed.

33:21Query number two was with the window

33:22function missing partition by. The rank

33:24has no partition by so it ranks all

33:26customers globally instead of resetting

33:27the rank within each tier.

33:29So, why it happens

33:32creates a single global window

33:34and we need the partition by.

33:37We have the corrected code which you can

33:39see the partition by is in over here.

33:42And then it says add partition by over

33:44the over clause. And then query three

33:46month-over-month growth wrong order by

33:47and lag.

33:50Again, why it happens specifically

33:53it has the buggy versus the correct

33:55results.

33:56And then our corrected query down below.

33:58And then it says summary the three bug

34:00patterns and then we have all this

34:02information. Fantastic. Obviously we

34:05would want to test this out as well just

34:07to make sure this specifically works.

34:09But I'm just going to assume it works

34:11for this particular video.

34:12All right, our second one we're going to

34:14do performance optimization and

34:16rewriting.

34:18So, we can clear out of this chat or we

34:19can keep going forward.

34:21And we're just going to keep going

34:23forward.

34:25I'm going to write this out.

34:28So,

34:30our prompt is I have a query in slow

34:31query SQL that returns the right data,

34:33but it takes 4 plus minutes to execute.

34:35There's definitely something wrong with

34:36how it's written. Analyze the query tell

34:38me what's making it slow. Rewrite to

34:39make it much faster. Find the

34:41optimization if you can estimate how

34:42much faster new version should be. The

34:44query selects customers with specific

34:45criteria filters. Help me understand why

34:47it's slow and how to make it more

34:48efficient. Uh this goes back to my first

34:51job when I was writing queries as a tax

34:54data analyst. And some of the queries

34:56that I had to rewrite from our engineers

34:58were terrible. They were just doing

34:59select stars. But I assume this error is

35:01going to be much more than a select

35:03star. We're going to run it and to see

35:05what happens.

35:06All right, a little bit longer on this

35:07one,

35:08uh but it went through over here and was

35:11able to fix it. So, let's take a look at

35:13what happened.

35:15So, four anti-patterns.

35:17Correlated subquery with nested scalar

35:20subquery inside exists, the main killer.

35:22So, it talks about this over here. The

35:24exists correlated subquery runs once per

35:26customer inside it. The scalar subquery

35:28that the optimizer may re-execute every

35:30row.

35:31At this scale, a cubic blow-up. This

35:34alone explains 4 plus minutes. Not in

35:36with a subquery problem is not in cannot

35:38short-circuit. It must compare every

35:39candidate against full subquery.

35:42Redundant distinct in all three

35:43subqueries. Self-referencing subquery in

35:46the same table. The rewrite

35:51super clean code, in my personal

35:52opinion.

35:54But this says, wait, the region filter

35:55needs a cleaner expression. The original

35:58over here. Let me rewrite that. So, it

36:00self-corrected.

36:02Then we have optimization strategy.

36:04Original, rewritten, why it's faster,

36:07speed estimate. So, conservative

36:10estimate 20 to 50 times faster on a real

36:13production data set. Wow.

36:17And uh

36:18assuming that this is all correct. We We

36:21don't know. We haven't tested out the

36:22exact code. But it was able to go and

36:25specifically rewrite it. So,

36:28whether you need to rewrite code or you

36:30need to look for specific errors,

36:33Claude is pretty good at it and it will

36:35give you learning opportunities to kind

36:37of grow out your skills. And yeah, so

36:39that's two ways we can review our code.

36:41We talked about the specific errors. We

36:43sent it to the Claude code. It told us

36:44what was wrong and gave us some

36:46feedback. And then also kind of speeding

36:48up opportunities. So, this code just

36:51took forever to run and identified some

36:53of the bottlenecks and was able to

36:55resolve it for us.

36:57So, this next section is kind of a

36:58throwback to my first data analytics job

37:01because one of my first projects that I

37:03had to do was convert hardcoded queries

37:07into views.

37:09The person or the team before me ended

37:11up having a bunch of SQL queries and

37:14they worked, but they were all

37:16specifically in our reporting layer. And

37:18that's just not the best approach to

37:20have, you know, 30, 40, 50 lines of SQL

37:23code in a report. You should just have a

37:25select specific columns from a view.

37:28And then you manage your view

37:30outside of that specific reporting

37:31layer. So, it was my goal to take all

37:34those queries, extract them out, write a

37:36view, and then replace the view, and

37:38make sure that report works. So, what I

37:41want to do today with you guys is build

37:43out some views with the help of Claude

37:46code. So, in this essence,

37:50we're going to slightly change up things

37:52and let me show you over here. Okay. So,

37:56we have three different reoccurring

37:57queries

37:59and maybe these are in a reporting

38:00layer, maybe these are queries that a

38:02data analyst runs every single week I

38:05just to get some updates about the

38:07company.

38:08And obviously you could just

38:09copy and paste this query, you could

38:11change up specific dates, but it is much

38:14more efficient to turn these into views

38:16and just be done with them. So,

38:19and that is what we're going to do. I'm

38:21going to go over into our cloud code.

38:23I'm going to remove my head and we're

38:25going to turn these specifically into

38:27views. So, here's our prompt.

38:32I'm going to grab this. It's a little

38:33bit of a lengthier one.

38:36So, I'm saying I have three SQL queries

38:38I run every Monday morning. We have

38:40weekly active users, revenue versus

38:41target customer health score. The

38:43queries work, but they're messy.

38:45So, they talked about issues on the

38:46quality, which we've already covered

38:48kind of in the last session, but it's

38:49good to rebuild upon it. I'm asking,

38:51"Can you convert each query into a clean

38:53SQL view, replace hard-coded dates with

38:54dynamic ones with current date and date

38:56functions?" Man, there's so many times

38:59that I've taken a look at incorrect

39:00queries and people just put hard-coded

39:02dates. It's it's terrible. If you're a

39:04data analyst and you're relatively new,

39:06please do not do that.

39:07Um

39:08move repeated logic into the view

39:10definition once, add comments

39:12explaining, and then make sure the views

39:14are named clearly. The goal I should

39:16want to select star from weekly active

39:17users on any Monday and get the right

39:19data automatically. And also tell me

39:21what these queries before and after. So,

39:23pretty straightforward.

39:25We'll run that and uh we're back.

39:28All right, that actually wasn't too bad.

39:29It took about 2 minutes.

39:31And uh super manual task in the past, I

39:33can tell you that. Uh but regardless,

39:36we have query one weekly active users

39:38problem.

39:42Replace both over here runs once,

39:44meaning today minus 7 days.

39:46Query two.

39:49Fix and then query three.

39:54Structural fix across all three.

39:57When you do this on any Monday, I'll

39:59compute it at once.

40:01But I don't see these queries in over

40:04here also,

40:06unless it just fixed it in that file.

40:08Oh, I see where the queries are right

40:09now. So, we have this view over here.

40:15It says run anything over here, create

40:17or views. So we have weekly active

40:19users.

40:20Our first view,

40:22revenue versus target,

40:24and then our customer health.

40:27And we have all of our information over

40:29here correctly formatted. So, obviously

40:32we can expand upon this and talk a

40:34little bit more within the section, but

40:36you get the point. Very good at writing

40:38SQL code. All right, let's move into our

40:40eighth example.

40:41This section addresses one of the least

40:43fun but most critical tasks that

40:45analysts have, and that is going to be

40:47documenting things. Now, personally I

40:50work in the compliance space, so when we

40:53work with regulators, it is super

40:54important that every query is

40:56documented, all of our database is up to

40:59date. But, if you're outside of

41:01compliance or like risk and

41:03underwriting, sometimes people get a

41:05little bit lazy, and that's not really

41:07helpful, especially as you bring on new

41:10members to your team or people leave.

41:12It's really important that you have some

41:14sort of documentation, otherwise people

41:16are going to become lost and hours are

41:18spent. So, we're going to take a look at

41:20two different things. Uh the first one

41:21is going to be undocumented code. You

41:23can think of this as like a script

41:24written months ago that has no comments

41:26and variable names that mean nothing.

41:28The second one is undocumented schemas.

41:31So, database tables with cryptic column

41:33names that require reverse engineering.

41:36Cloud Code can solve both of these

41:37situations automatically, transforming

41:40messy code into clear, maintainable

41:42documentation and building data

41:43dictionaries from just raw schemas. So,

41:46we're going to go through two different

41:47prompts and jump into that.

41:50Okay. So, I've already uh removed my

41:52head because I don't want to get in the

41:54way. Uh but you can see we have like the

41:56schema dump over here.

41:58And then we also have this Python

42:00script,

42:01which really means nothing. We have no

42:03comments, nothing across the board. It's

42:04pretty bad. So, we're going to go and

42:07fix both of those. So, our first one is

42:09going to be the Python script

42:11documentation.

42:13And here is our query.

42:15So, it says I wrote this Python script a

42:17few months ago and honestly I don't

42:19remember exactly it does. Can you and it

42:21says read through it and explain in

42:22plain English what it does, rename all

42:24the variables to descriptive, add proper

42:26comments, write a comprehensive

42:28docstring at the top. The script should

42:29be production-ready when you are done.

42:32And maybe I want to put our exact script

42:34over here, which is called undocumented

42:36script.py.

42:38It's Python script.

42:41Undocumented

42:44script.py.

42:46Cool.

42:47And let's run that. All right, and it's

42:49ready to go. Here's what the script does

42:51in plain English. So, it tells us the

42:53steps 1 through 6. A few issues I'm

42:55fixing as part of making it

42:57production-ready. So, it's actually

42:58fixing some of our code as well.

43:01Here's a summary of the changes made.

43:02So, plain English explanation, variables

43:05being renamed,

43:07bug fixes,

43:09and let's take a look at our

43:10undocumented script. So,

43:13we move this over.

43:17You guys can see over here now, we have

43:19our top section, which was not here

43:21initially.

43:23And then,

43:25query one,

43:27query two,

43:29a lot of notes

43:31re-labeled.

43:32And overall, I think this is a way

43:34better experience. Again, I'm just

43:36glossing through this,

43:38but if you are doing this at a company,

43:41you'd obviously want to double-check it.

43:42You would want to rewrite some stuff and

43:44rework it, but

43:46I think this is okay for this demo.

43:49Also, every company has their own style

43:51guides. So, maybe you would want to

43:52build out a style guide associated with

43:54it on what needs to be documented or

43:56what not.

43:58Outside the scope of this particular

43:59video, just kind of showing you we can

44:01document this. Okay, let's jump into our

44:04second prompt now. I'm just going to

44:07make this full screen.

44:08And our second one is going to be the

44:09data dictionary. So, I'll get out of the

44:11way for you guys.

44:13And

44:14let's write that out.

44:18Cool. So, I also have a database dump an

44:21engineer gave me and it has a bunch of

44:22tables with cryptic column names. Can

44:24you build a data dictionary with

44:25markdown table with these columns? Make

44:27reasonable inferences about what type of

44:29cryptic columns mean based off the names

44:31and the tables that they're in. Assume

44:32this is a SaaS analytics database.

44:36And we'll be back when it's ready. Okay.

44:38And uh we're done. So, I wrote a

44:40dictionary which will cover this in a

44:42second. It talks about the columns. So,

44:45decoded over here for each of these

44:47columns. So, you can see cost,

44:48acquisition source,

44:51monthly recurring revenue, a standard

44:53SaaS metric, subscription tier pricing.

44:56Few things worth noting, like an

44:58engineer, days of week in dim dates has

45:00no documented convention,

45:03no document scale at the schema,

45:05ambiguous whether this is pre or post

45:06discount. And if we jump into our data

45:09dictionary

45:11and scroll this all the way over,

45:16this looks pretty good.

45:19So,

45:21very great job

45:24documenting

45:26database as well as that particular

45:29Python query.

45:31So, this section, what we're going to do

45:32is create a mock-up dashboard. Now, in

45:36the real world, we're probably going to

45:37store all your dashboards in like a

45:39Power BI or a Tableau or a Metabase or

45:42whatever BI reporting layer you that you

45:44want, but sometimes you're going to have

45:46data in a spreadsheet and you need to

45:48just throw something together really

45:49fast and Cloud Code can essentially

45:52build us out a HTML dashboard within

45:55minutes, something that would take hours

45:57to build in another platform. So, let me

46:00show you the spreadsheet we're going to

46:02use and then let's build this out

46:03together.

46:04So, very similar

46:08very similar data that we've been

46:09covering so far throughout this video.

46:11Month, region, product, revenue, new

46:13customers, churn customers, support

46:15tickets, NPS score, ad spend, as well as

46:18a sales rep counts. And let's jump into

46:22Cloud Code.

46:24And we'll get out of your guys' way.

46:25And let's expand that out. Cool. So,

46:28here is

46:30our query we're going to write.

46:35I have a dashboard data CSV with 2 years

46:37of monthly business metrics across four

46:39regions and three product line. Build me

46:40HTML dashboard with the following

46:41visualizations: revenue trend, new

46:44versus churn, NPS score over time,

46:46summary scorecard.

46:47I make it look professional, something I

46:49could show my leadership team, use a

46:50clean color scheme, responsive layout,

46:52make it something I can open in any

46:54browser. Now, obviously, you can define

46:57colors, you could add in a lot more

46:58details over here, but for demo

47:00purposes, let's give it a shot.

47:03All right, it says our dashboard is

47:05ready to go. This took me

47:084 minutes,

47:09give or take.

47:11And

47:12we'll take a look at that here in a

47:13second. We have total revenue, new

47:14customers, average NPS, three charts,

47:17notable story in the data dashboard

47:18visible at the top of our cell, strong

47:20growth through mid-2024 followed by a

47:22concerning reversal.

47:23And let's take a look at what this looks

47:25like.

47:26Okay. So, this is our dashboard. Again,

47:28you can change all the different colors,

47:29you can make this look way nicer as you

47:33want.

47:34And uh we have our total revenue, new

47:36customers, churn rate, NPS score. I

47:38think overall, I think this looks very

47:40clean.

47:41Uh we have revenue by region. You can

47:43see we can highlight this.

47:48North America, Europe, Asia, Latin

47:50America.

47:52New churn customers, you can see

47:55we're gaining a lot of new customers,

47:57but churn is also increasing quite a

47:59bit. Then we have NPS score over time,

48:02which went up, flatlined, and then it

48:04really went down. Now, obviously,

48:07over here, this points to a pretty clear

48:09picture, but not always the case in

48:12real-world data. So, that is an HTML

48:15dashboard design,

48:18and uh next section we're going to cover

48:20Streamlit, and I will see you there.

48:23So, for section number 10, we're going

48:24to be covering something called

48:26Streamlit. Streamlit's honestly one of

48:27my favorite Python libraries, and it

48:29allows us to quickly generate data apps

48:32that we could share with our teams. At

48:34my last job, I built 15 different data

48:36apps that probably saved like anywhere

48:38from 30 to 40 hours every single week,

48:40and that stopped from other people in my

48:43team asking me, "Hey, can you help me

48:44clean up this spreadsheet, or can you

48:46combine these spreadsheets together and

48:49throw this together, or hey, I have this

48:50external data source, can you help me

48:52plot all of this data?" I would just

48:54build them Streamlit apps, and then

48:56teach them how to use Streamlit just

48:57dragging and dropping in a spreadsheet,

48:59and people were self-sufficient. So,

49:02I think it's a very powerful skill for

49:04you to pick up as a data analyst,

49:05especially if you know the basics of

49:07Python. Really spend some time with

49:09Streamlit. Let me just show you the

49:10website really fast as well.

49:12And obviously, this is not a Streamlit

49:14tutorial. I just kind of want to show

49:17you that this is out there. So, faster

49:18way to build and share data apps, and

49:21you can take a look at the gallery and

49:23different things like people build out.

49:25So, you can see dashboard designs,

49:27which is great. You can see overviews.

49:31Um

49:32a lot of people will build out some

49:34stuff around machine learning also with

49:36Streamlit,

49:38and data science.

49:40You can see this is like a Streamlit

49:41cheat sheet that someone built out. But

49:43these are just like very quick web apps,

49:45and I would say like this is a step

49:46above the HTML ones, because you need to

49:48host these. Uh Uh can host them free in

49:51Streamlit Cloud, or you could host them

49:53privately. Uh so, like if you have a AWS

49:56instance, your data engineering team can

49:57help you set up all to host these

49:59internally.

50:01There's also ways uh to host this if you

50:03have like a Snowflake subscription.

50:05Again, not the purpose of the video

50:06isn't fully to demonstrate everything

50:09about Streamlit and how it works. Just

50:11showing you that you can build out quick

50:12data apps with Python with Streamlit,

50:15and it really should replace those HTML

50:17dashboards, which I just showed you in

50:18that last section. In addition, this is

50:21really good if you want to combine

50:22multiple data sources. Like imagine you

50:24have a data source in your database, you

50:26have an external data source from a

50:28vendor, and then you have like a third

50:29data source somewhere else in a

50:31spreadsheet, and you have to constantly

50:32combine them together. Uh you can build

50:35out a Streamlit tool to do so. Okay.

50:38So, what we're going to do is build out

50:40a Streamlit app for this over here.

50:43And this is just another spreadsheet,

50:45and maybe we just get questions over and

50:47over again from our internal team. So,

50:50we want to automate this and build out a

50:52streamlit.py file. So,

50:55I have our data over here, and let me

50:58just walk you through these prompts

51:00really fast.

51:01So, we have two different prompts.

51:05And I will walk you through this. Cool.

51:07So, the first one it says, "I have a

51:08customer CSV with customer information

51:10from all accounts. My team is constantly

51:12asking me to look up customer details.

51:14Instead of being the bottleneck, I want

51:15them to

51:16build a Streamlit app where they can

51:18search,

51:19filter, have a full table view, make it

51:21professional. My sales team should be

51:23able to use it immediately. Save it as

51:26this Python file."

51:28Again, you're going to have to host this

51:29Python file. It's only going to be

51:31available locally, but I will show you

51:33really fast how we can deploy this

51:35within Streamlit Cloud. So, we're going

51:37to run this, and we have our results

51:40over here. So, customer lookup.py ready,

51:43run it over here. I'm not going to run

51:44this now. I'm actually going to deploy

51:46this in GitHub and just show you how to

51:47do that really fast.

51:49But, I'm also going to ask for a second

51:51section of this.

51:53And let's imagine we've re-prompted this

51:55again.

51:56As I mentioned, you're going to

51:57constantly want to re-prompt. So, I'm

51:59going to say, "Great, the team loves it,

52:00but now I want to add a second tab that

52:02shows management insights. Add a

52:04dashboard tab to the app that displays

52:05metrics, total customers, total

52:07contract, and then some charts, and make

52:10it visually distinct."

52:14So, we're going to upload that over

52:15here.

52:20And also, while that loads up,

52:22just to show you,

52:24we have this over here. Now, I'm on a

52:27relatively new computer, so I haven't

52:30installed Streamlit on this one as of

52:32yet. I need to still install a bunch of

52:34different Python libraries.

52:36But, you can see that we have our

52:38imports here at the top. We have a page

52:40config, which is helpful on there.

52:43We have caching, which is a typical

52:45issue people have with Streamlit.

52:47And we have this whole Streamlit code

52:50over here.

52:51All right. I'm just going to keep

52:53allowing it to update this in real time.

52:57And we have different tabs as well.

52:59Just based off of, you know, the

53:00conversation we're having right now.

53:03I'm going to keep letting it update our

53:04code.

53:06Looks pretty nice across the board.

53:08You have our tab over here as well. And

53:10you know, in the past, I had to write

53:12all this out manually.

53:16In fact, we have like a 4-hour Python

53:18Streamlit course here on the channel.

53:21And you know, there's a lot of benefits

53:22if you want to learn the basics, but a

53:24lot of this is now automated. Okay. So,

53:29it's

53:31should be good to go.

53:33I'm going to close this out, and you can

53:35see it generated all of this code

53:38for us, about 300 lines.

53:41And sometimes these Streamlit

53:43codes can be a little bit longer.

53:44Uh you can see I also don't have Plotly

53:46on this computer. That's okay.

53:49Yeah, so we can also upload this

53:50directly into Streamlit.

53:54And make sure that you have a Streamlit

53:56account. It is free. So you can see

53:57deploying free. There is pro versions

53:59also.

54:00Uh so you can deploy with Snowflake. And

54:02you have a trial. This is just public

54:04app. So we're going to do free.

54:06I'm going to sign in.

54:10Sign in with Google.

54:12Okay. Then you go over here to the top

54:14right where it says create app.

54:16And then I'm going to say deploy a

54:18public app from GitHub.

54:21And I need to upload this into GitHub

54:22now.

54:23So I'm going to go and create a new

54:24repository.

54:27And what I'm going to call this over

54:29here is

54:30cloud

54:33code Streamlit.

54:37demo

54:40I'm just going to click create

54:41repository.

54:44And then we're going to add this into

54:45our repository. I'm just going to create

54:47our main file. I'm going to call this

54:49main.py.

54:51Like that.

54:52And I'm going to copy over our code.

54:55Obviously there's more efficient ways to

54:57upload stuff, but I just want to show

54:59you a very easy way that we could do

55:00that.

55:02We're going to commit our changes.

55:07Okay.

55:08We need to add in our other file as

55:09well. So we're going upload our

55:11spreadsheet. Let me just grab that from

55:13the folder.

55:15Great. So we have our customer data as

55:16well as our main.py.

55:18And then I'm just going to go over here

55:20and look for this.

55:22Then I go back over here. Just grab this

55:23cloud code demo.

55:25main branch

55:27We're going to grab main.py.

55:29And I'm going to deploy this.

55:32And as it deploys it's going to say your

55:34app is in the oven.

55:36And once again, this is public. So,

55:39anyone can see the code associated with

55:41it. Anyone can specifically see this

55:44specific app. So,

55:45if you have

55:48stuff out there that you don't want

55:49people to see,

55:51if you have private stuff out there that

55:52you don't want people to see, not the

55:53approach. Okay. And then I have an

55:55error. So,

55:56the reason why is you can see this

55:58import plotly, and you have to import

56:01that in with Streamlit. So, I'm just

56:03going to pause and get that in really

56:05fast. Okay. And now it's working. All I

56:07did is just add a quick

56:08requirements.text file and threw in the

56:11Python library plotly.

56:13And you can see now we have a dashboard

56:15with all of our data.

56:17And this is all obviously contingent on

56:20that spreadsheet that we've uploaded in

56:22over here,

56:23which is this customer data.

56:25And you can see our dashboard has been

56:27created.

56:31yellow, red.

56:33And then we could also go back into this

56:34customer lookup and then search by

56:36company name.

56:38So,

56:39or ID. Let's say I just want to go over

56:41here and say C002. So, we'll say C002.

56:46And we have one result.

56:48And the information associated with that

56:50specific company. Again, this literally

56:52took maybe like 5 to 10 minutes in

56:55general between just prompting and

56:57uploading this into GitHub.

56:59And now we have something that our team

57:02can just go through, search for a

57:04company, or search

57:06uh for that ID and get some specific

57:08information. Obviously, this is

57:10contingent on a hard-coded CSV file,

57:13which is not the best approach, but

57:15a very quick demo for YouTube purposes.

57:17Hopefully, this gets the wheels spinning

57:19and you're like, "Oh, I have a ton of

57:20different ideas." And yes, uh if you

57:23build this out correctly, you can have

57:24Streamlit talk to your database. Being a

57:27public app on Streamlit Cloud,

57:30do not do that. Please. That is a huge

57:32security flag. If it's privately hosted

57:34in Snowflake or AWS, go for it, but uh

57:37public like I just threw over here, do

57:39not do that. Remember, anyone can access

57:42uh your code because this has to be

57:44public

57:46for Streamlit Cloud to work if you're

57:48having it as a free deployment. So,

57:51anyone could have this.

57:52You could also just run Streamlit apps

57:55privately if you specifically want to do

57:57so, but again,

57:59two prompts, we were able to build out

58:01this customer hub, and I think this is

58:04pretty awesome. So, those are 10

58:05different ways that you can become a

58:07more efficient data analyst with the

58:09help of Cloud Code. Hopefully, you found

58:12a little bit of value in this video. If

58:14you did, make sure to subscribe it to

58:15the channel, and if you're interested in

58:17a part two, let me know down below.

58:20Honestly, I have like 20 or 30 different

58:21ideas listed in a doc, and would love to

58:24make a follow-up video. If there's a

58:26technique that you are also utilizing

58:28that I didn't cover in this video, feel

58:30free to leave it as a comment, and maybe

58:31it'll be one of those featured in that

58:33next video. All right, I'll see you guys

58:35in another one.