0:00

We all know that tools

0:02

like ChatTPT have really empowered developers

0:04

to tackle bigger problems. Are

0:07

using Tailwind CSS and need a login

0:09

page? Try asking chat what is the

0:11

HTML for a login page with a

0:14

login, username, password, and a button in

0:16

its own section in the center of a page. It

0:19

will literally give you the first pass version

0:21

of it. But how far can we

0:23

push this? Fred to be a lot may

0:26

have taken it farther than most. He

0:29

built a functioning SaaS product

0:31

with paying customers by using

0:33

only ChatTPT and Python. It's

0:36

fascinating to hear his story and see what he's built. This

0:39

is TalkPythonMe, episode 459, recorded February 19th, 2024. Welcome

0:58

to TalkPythonMe, a weekly podcast on

1:00

Python. This is your host, Michael

1:02

Kennedy. Follow me on Mastodon, where

1:05

I'm at mkennedy and follow the

1:07

podcast using at TalkPython, both on

1:09

bostodon.org. Keep up with

1:11

the show and listen to over seven years

1:13

of past episodes at talkpython.fm. We've

1:16

started streaming most of our episodes live

1:19

on YouTube. Subscribe to our YouTube channel

1:21

over at talkpython.fm slash YouTube to get

1:23

notified about upcoming shows and be part

1:25

of that episode. This

1:28

episode is sponsored by MailTrap, an

1:31

email delivery platform that developers love.

1:33

Try for free at mailtrap.io.

1:37

And it's brought to you by us over

1:39

at TalkPython Training. Did you know that

1:41

we have over 250 hours of Python courses? Yeah,

1:44

that's right. Check them out

1:46

at talkpython.fm slash courses. Fred,

1:50

welcome to TalkPythonMe. So great to be

1:52

here. Thank you so much for the

1:54

invitation. Nice to be here. Yeah,

1:56

it's lovely to have you. We're going to have

1:58

a lot of fun talking. about this

2:00

app that you've built using

2:03

highly powered by AI in

2:06

different ways. It's both powered by

2:08

AI, but also largely built with

2:10

AI. And I think it's going

2:12

to give people an interesting look

2:14

into what is possible these days.

2:17

What is possible and the journey of a middle-aged

2:19

coder or so maybe. That's really an interesting

2:22

thing to discuss. And maybe we could

2:24

just start a little bit with the

2:26

introduction by yourself and then talk a

2:28

bit about that. So before we dive into

2:30

talking about how you use chat

2:32

GPT to largely build this app

2:34

that is a fast product service

2:36

on the internet people pay subscribe

2:38

to, which you know is the

2:41

dream of many people is awesome

2:43

with AI. You know, just tell people a

2:45

bit about yourself and then we'll I want to talk about this, this

2:48

journey. It's a long story because I'm

2:50

turning 15 in June, 2024. So

2:54

if you listen to this before, June, 2024, I will be 49 after

2:56

I would be 50. So

3:00

yes, that will be your spring chicken.

3:03

I'm older than you, Fred. I

3:08

found someone older than me to talk to.

3:10

Oh, that's fine. You can still write code

3:12

into your 50s. Let me tell you the

3:14

keyboard store. It's amazing. And I

3:16

started when I was 48. Maybe we get back

3:18

to this later. So

3:20

originally, my professional

3:22

journey started as an interpreter, Russian,

3:25

English to French. So I

3:27

graduated as an interpreter. So

3:29

in languages. And

3:31

then I started being involved in music.

3:34

I was head of sponsorship for a

3:36

big music festival in Belgium. I had

3:38

rock venues like music venues for a

3:40

few years. And then

3:42

I started being involved in tech as head

3:44

of product head of paid acquisition

3:46

for a few startups. I had

3:49

my own startups. I existed one

3:51

startup in the FinTech space in

3:53

the early 2000s. I

3:55

had a circus for two years. That's

3:58

quite interesting. story. Yes,

4:01

family circuit with four big tents, eight

4:03

tons of equipment. So that was, that

4:06

was quite a story. Was this like

4:08

Cirque du Soleil? Or was it with

4:10

animals? Or what kind of? It was

4:13

more like, I would say, like Cirque

4:15

du Soleil, but at a smaller, smaller

4:17

scale, of course. Okay. But four tents

4:20

and circus artists and actors and musicians

4:22

and that was an original show, an

4:24

original story. I also produced rock bands.

4:27

Maybe I can give a shout out

4:29

to one of the bands that

4:31

I love that I contributed to,

4:33

the Boxer Rebellion. Maybe you heard

4:35

about them, the Boxer Rebellion. Yeah.

4:37

It's a British band based here

4:39

in London, but with a singer

4:41

from from Nashville, Tennessee, Nathan Nicholson.

4:44

So I was involved in music in

4:46

tech, I mean, in all of those

4:48

things. And I think the project that

4:50

we're going to talk about is kind

4:52

of a good mix between music, tech

4:54

and all my passions. I was also

4:56

radio DJ for you. So that's why.

4:58

That's the audio side, right? That

5:00

kind of brings you some together in the music,

5:02

I suppose, as well. Exactly. Amazing. You know, listening

5:04

to tell the story, first of all, what a

5:07

diverse set of experiences

5:09

that must have been really fun. It's

5:11

not a career. That's the problem. Unemployable.

5:15

Yeah, that's, but in terms of thinking

5:17

about products and coding, I know when I

5:19

experience different areas, I always see like, oh,

5:21

this is could be way better. And that

5:24

could be way better. And why is that

5:26

so hard? Surely we could create some software

5:28

to fix these things. You must have had

5:30

a lot of ideas about programs you would

5:33

like to see and stuff that you'd like

5:35

to work on. And multiple times, the solution

5:37

for me was to hire people to do

5:39

it. I had my own startups in the

5:42

past. And I always had a team of,

5:44

let's say, two to five developers working for

5:46

me when we were doing that. So

5:48

I kind of understood all the structure, the

5:51

architecture of the apps. But

5:54

I'd never really coded myself. So

5:56

I was always relying on other

5:58

people. The difference to The day is

6:00

that I spent most of my days coding,

6:02

which is quite different. Yeah, and you said

6:05

you started coding when you were 48. Yes.

6:09

And roughly almost two years ago.

6:11

Yes. That's awesome. You know,

6:13

I do hear from a lot of people or see

6:15

comments like, oh, I think I'm too old to learn

6:17

this or, oh, you have

6:19

to have done this when you were a

6:21

kid or maybe gotten a degree in college.

6:24

Otherwise, you're doomed. You know, you surely have

6:26

an alternative viewpoint on that, given your success.

6:28

I thought it would be

6:30

like that also for me because I

6:32

would say I'm quite a perfectionist. So

6:35

if I learn something, I want to

6:37

achieve something with what I learn. And

6:40

the problem, the issue with code for me

6:42

for a long time had been that

6:44

I could go through the

6:47

basic tutorials, the first lessons, you

6:49

know, watching some videos or reading

6:52

some tutorials. It was difficult

6:54

for me to really achieve something meaningful,

6:56

something that I could really share that

6:59

could become a product because

7:01

you always end up doing the

7:03

same little exercises and little tutorials.

7:06

And that doesn't end up anywhere. You learned

7:08

how to do a loop, but then that

7:10

doesn't really help you get any closer to

7:12

an app, right? To do a little to-do

7:14

list or a little calculator or, you know,

7:17

something very simple, nothing that that would be

7:19

worth sharing as a product and especially as

7:21

a paid product. So I always thought, okay,

7:23

if I want to build a SaaS, the

7:25

only solution for me is to rely on

7:28

a freelancer or on an agency

7:31

or on a team of people. I

7:33

was doing it until I discovered. That's

7:35

a big risk. And it's also not

7:38

that easy to hire a team to

7:40

build your product when you're still trying

7:42

to figure out what a product is

7:44

and all these things, right? Even if

7:47

you hire non-expensive developers, it

7:49

has still a cost. And if

7:51

you want to, you know, just if

7:53

you don't want to take too much risk, if you want

7:56

to test your product to see if you have product market fits,

7:58

I mean, you don't want to do want to

8:00

spend 5,000, 10,000, 15,000

8:03

on a product. It's difficult. And

8:05

the alternative that people are most

8:07

of the time advocating these days

8:09

is no code. So it's building

8:11

stuff with a, let's say, bubble,

8:14

which is pretty popular. Got retool.

8:16

They've been a sponsor of the

8:18

show before. Retool, obviously,

8:21

that's pretty nice, mostly for

8:23

internal apps. And also you

8:25

can just connect stuff through

8:27

Zapier, through Make, formerly Integromat.

8:30

You can achieve something, but you

8:33

don't have the control on the UI.

8:35

You don't really have the control exactly

8:37

on what's happening behind the scenes. And

8:39

you don't have the flexibility that you

8:41

have if you really code yourself. So

8:43

you can really get your hands dirty

8:45

and code on your own. So I

8:47

used also to be a strong advocate

8:50

of no code. So I was using,

8:52

I'm still using firstly Zapier and then

8:54

later on Make. But

8:56

once I discovered Python

8:58

in August, September, 2022,

9:01

then everything changed for me

9:03

because then I could really

9:06

start doing my own stuff.

9:08

And I started with a

9:10

backend stuff with a backend

9:13

scripts to create competitive intelligence

9:15

scripts, things like that. And

9:17

I slowly moved into frontend.

9:20

But first for a few

9:22

months, it was mostly local

9:25

scripts on my computer. That's really interesting.

9:27

The low code, no code stuff is

9:30

super powerful. I think it does make it

9:32

possible for people to build apps.

9:34

I have a friend who built a

9:37

platform a little bit on top of something

9:39

that's kind of a low code thing. And

9:41

it's what they're doing is really impressive. And

9:43

they've got lots of customers and it's going

9:46

really well. It's a vibrant business. It's always

9:48

your beholden to whatever that platform

9:51

does. The platform has downtime, you

9:53

have downtime. If they're changing

9:55

their feature set in a way that doesn't quite

9:57

line up with what you're trying to do. That's

10:00

too bad because, you know, that they've

10:02

changed their, their direction or whatever. And

10:04

it's, it's really powerful, but it's also

10:06

a little bit of a risk, right?

10:08

To be built on top of that.

10:10

A risk, yes. But for me, it

10:12

was even more, I would say, the

10:15

arbitrage was even more about flexibility. So

10:17

when you code something on

10:19

your own with Python, JavaScript, as

10:21

I'm doing now, it's faster in

10:23

many cases than trying to figure

10:25

out how to connect, let's say,

10:27

air table with something else, then

10:29

you've got to think about how the

10:31

data is structured in air table, how the

10:33

data has to be structured on the other

10:36

side, and then how to connect

10:38

them together. Sometimes it's

10:40

easier if you just have

10:42

a SQLite database or MySQL

10:44

database, and then if you

10:46

do your own stuff on

10:49

your own without trying to

10:51

connect existing SaaS

10:54

together. It was more about flexibility

10:56

for me. The main reason why

10:58

I switched almost completely from a

11:00

no code to a coding. That's really interesting.

11:02

I guess one more that would be worth

11:05

considering is price. Some of these low code,

11:07

no code platforms are really powerful, but when

11:09

you start to look at price for a

11:11

decent number of users using your app, a

11:13

lot of them are built for internal use

11:15

cases that are not really built for having

11:18

a bunch of multi-tenant style users and

11:20

so on. But also they can just

11:22

get real expensive real quick. Yes, because

11:25

they are based usually on the amount

11:27

of transactions or

11:29

scenarios or zaps that can

11:31

run. So if you want to

11:34

transfer one piece of data from

11:36

one software to another one, it's

11:38

one transaction. If you have, let's say,

11:41

10,000 transactions per month for a certain

11:43

price, then you can run out of

11:45

transactions pretty quickly. And very

11:48

quickly, you end up paying

11:51

25, 30, 40 dollars

11:53

for a pretty low amount of

11:55

transactions, API codes basically between the

11:58

different SaaS. And if you

12:00

glue them together, one

12:02

tool connects to another, which connects to another, then

12:04

you've got to pay for kind of the traffic

12:07

and the activity in all three. It's not just

12:09

like, well, I'm talking to my database. I'm now

12:11

talking to some other service that's got something

12:14

going on. And the fact is also that

12:16

there are a few tools that enable you

12:18

to create a UI user interface for your

12:20

product, but you're limited to the building blocks

12:22

usually that they offer, even if you can

12:25

tweak them with some code. And

12:27

it's a bit the same. I would say it's also

12:29

the same issue that I've experienced

12:31

on WordPress. The

12:33

problem is that if you use something

12:35

like bubble and you want to build

12:37

something with it, you bring to your

12:40

app or the mothership, the core of

12:42

the big app of bubble as a

12:44

whole, when you want to build something

12:46

on top of a platform like that.

12:48

It's a lot of things that you

12:50

don't really need for your specific apps,

12:52

but that has to be there because

12:54

the whole engine has to run. That's

12:57

usually what you also experience with

12:59

WordPress. For instance, when you have a

13:02

site on WordPress, maybe sometimes or even

13:04

most of the time you don't need

13:06

that big engine. You could

13:08

work with a static website with a

13:11

few HTML, CSS, JavaScript pages. I would

13:13

be much lighter. There would be less

13:15

code to activate to have your website

13:17

up and running. So in terms of

13:20

performance, if you can code your own

13:22

stuff, I think it's also a good

13:24

reason to choose coding

13:26

versus no code if you can do

13:28

it. Totally agree. Price, performance,

13:31

control, all those things. That

13:33

said, you do have to understand more

13:36

about the world. One thing

13:38

we were coordinating around is just

13:40

understanding web app architecture.

13:43

I don't know how you feel about it,

13:45

but I feel like a lot of the

13:47

guidance that is out there is too intense.

13:50

There's too many moving parts. Everyone

13:52

thinks you need to have some kind

13:54

of background worker service like Celery, and

13:57

you need a caching layer, and

13:59

you need this. and you need that and

14:01

it just starts to be like, whoa,

14:03

this is like running Netflix. You

14:05

know what I mean? Whereas a lot of

14:08

times what you need is a backend, a

14:10

frontend and some kind of data store and

14:12

that's it. And that data store, you mentioned

14:14

it before, the data store can be in

14:17

the early days, could be SQLite, which doesn't

14:19

require another server or another thing to maintain

14:21

and connect and secure and it can just

14:23

be a file that you back up. Exactly.

14:25

While you're proving out the idea, right? So

14:28

maybe you're coming from a place where you

14:30

weren't a DevOps expert. Like

14:33

how does transition go for

14:35

you? When I started developing

14:37

an app, for some reason,

14:39

I came across Flask as

14:41

a solution for Python. So

14:43

I started developing the app

14:45

on Flask. And what I

14:47

like about Flask is that

14:49

if you compare it, let's

14:52

say you use Flask and then for

14:54

the pages, you use HTML, CSS and

14:57

Vania JavaScript. What I like is that

14:59

you can deploy an app which

15:01

comes from a single folder and

15:03

you've got everything there, your main

15:06

Flask file and then all the, let's

15:08

say the sub files in Python with

15:10

all the business logic that you can

15:13

isolate in one file or another. It's

15:16

one project and then you deploy the

15:18

project to GitHub and then let's say

15:20

you connect GitHub to Railway. I'm using

15:23

Railway for my deployment. So you connect

15:25

GitHub to Railway and

15:27

then once you update

15:29

your code in PyCharm or

15:32

in Visual Studio, then you have the

15:35

automatic deployment to Railway. But it's one

15:37

project. If you want to decouple completely

15:39

the backend from the front end, let's

15:41

say you work with something like Next.js,

15:43

for instance, and then you want to

15:46

have a backend into something else, then

15:49

you've got to have multiple instances and

15:51

connect them together. It becomes

15:53

more complicated. But if you want

15:55

to create something simple, you can

15:57

keep it simple with one single

15:59

flap. web app and that's

16:01

what I've been doing so far. Most

16:04

of the iJingle maker that I

16:06

built, that I've been building since

16:08

September 23, is a Flask web

16:10

app. So it's one project, one

16:13

big project, it's one project. It's interesting

16:15

that you mentioned Celery a few moments

16:17

ago because I'm about to ship a

16:19

new app which is

16:21

coming in the next two weeks or

16:23

three weeks I think and that's the

16:26

first time after almost 18 months of

16:29

development that I've considered using

16:31

something like that because I needed it

16:33

for performance reasons to process something. Something

16:35

you want to start off in the

16:37

background and get back and just let

16:39

it be done later right? Yeah. And

16:42

I was not that easy. We had

16:44

a discussion with someone

16:46

who is amazing on the discord

16:48

of a railway two or three

16:50

days ago because

16:52

if you're deploying things on

16:54

railway you would usually use and

16:57

that's what they advise you to do

16:59

as on Iroku a proc file and

17:01

the proc file you've got the

17:04

start commands that will start your

17:06

project but on Iroku

17:08

which is a competitor of railway

17:10

you can start both the

17:13

web app and the worker

17:15

in the same file but you can't

17:17

do that on railway. You've got theoretically

17:19

two different projects which kind of defeats

17:21

the purpose. You kind of deploy them

17:23

both right but then then you get

17:25

to more coordination. Yeah but fortunately and

17:28

there is a discussion on the discord

17:30

between me and one of their guys

17:32

there they had a way

17:34

to create a custom JSON for

17:36

me which managed to start

17:38

both the web app and the worker

17:41

in the same in the same deployment

17:44

but I had to fight to be

17:46

honest with you for like four or

17:48

five hours to find a solution because

17:50

sometimes the problem and that's the issue

17:52

that I've seen a lot is that

17:55

traditional developers like professional developers not people

17:57

like me not random guys like me

17:59

usually they assume that the

18:01

people that they're supposed to explain

18:03

things to know much more than

18:06

what they do. And

18:08

so they just write a few

18:10

instructions, you know, like cryptic instructions

18:12

that nobody really understands. And they

18:15

would assume that everyone reading those

18:17

instructions would understand what they have

18:19

to do to, let's say, use

18:21

this worker service, for instance, but

18:24

that's not the case. I totally

18:26

agree. This

18:29

portion of TalkBython.me is brought to

18:31

you by Mailtrap, an email delivery

18:33

platform that developers love. An

18:36

email sending solution with industry-best

18:38

analytics, SMTP and email

18:41

API SDKs for major programming

18:43

languages, and 24-7 human

18:46

support. Try for free

18:48

at mailtrap.io. You

18:51

know, it's kind of the curse of knowledge. Like once

18:53

you know all of these things, it's

18:55

really hard to remember, oh, yeah, so and you

18:58

just, oh, just run that in Docker, and then

19:00

link it and set up a network over there.

19:02

Like, wait a minute. Yeah, I'm just trying to

19:04

run my

19:06

Python app, like you're bringing a whole new

19:08

thing in that you think makes it sound

19:10

easy. And it makes it sound real hard

19:12

to me. I totally agree. This is a

19:14

problem. So many moving pieces when you when

19:16

you start coding, I mean, obviously,

19:18

you know that because you've been in this

19:20

game for much longer than me. But

19:23

once you start coding, there are so many little

19:25

things that you have to know to make this

19:27

thing work. So it starts with

19:30

how to just configure GitHub and connect

19:32

it to your How

19:34

do you get if you have a

19:36

Git conflict, you might be like, well,

19:38

this is done. I can't do anything

19:40

else anymore. It's broken. Yeah, I mean,

19:42

it's really overwhelming when you're new, right?

19:44

There's a lot. And that's where AI

19:46

today helps you. Because in most cases,

19:48

if you know how to talk with

19:51

the AI, you will get answers

19:53

that are much easier to understand that

19:55

what you would have on Stack Overflow,

19:57

for instance, I agree. It's very interesting

19:59

how helpful it is. I think it's

20:01

a good time to about jump into

20:03

that. But before we do, tell people

20:05

about Railway a bit. How does this

20:07

work? I haven't used Railways before. It's

20:09

a platform where you can deploy your

20:11

apps, basically. It connects to your GitHub

20:13

account. It's

20:17

not only, I mean, you have

20:19

on Railway not only a possibility

20:21

to deploy your Flashcap or other

20:24

types of apps also, but they

20:26

also offer all the database aspects.

20:28

So you can spin

20:31

a Redis database. You

20:33

can have a MySQL database. I think

20:35

they also have Mongo. I mean, they

20:37

have most of the formats that

20:39

you can expect. And the great thing

20:41

is that everything is on

20:43

the same platform. Let's say that I've

20:46

got another project I'm involved with, with

20:48

a developer, with a real developer, I2Alex,

20:50

if you listen to this podcast. A

20:53

professional developer. For that

20:55

project, for instance, we have an

20:57

app on Eroku. We have a

21:00

database on MongoDB Atlas.

21:02

And we have the

21:05

files uploaded

21:08

to S3, to Amazon S3, for the

21:10

file storage. On Railway, you

21:12

can, at least it's what I do, you can have

21:15

at the same time, the deployment of

21:17

the app, which replaces Eroku. And

21:19

you can have the database also together. And

21:22

the cool thing is that they are, I

21:24

would say even physically at

21:26

the same location. So in terms of

21:28

performance, it's pretty easy to handle.

21:31

And very user friendly, I mean,

21:33

very, very easy to use from

21:35

what I saw even easier to

21:37

understand than Eroku. And

21:39

I tried also some things like Versal

21:42

or I use for some project also,

21:44

but that's more for static website Netlify.

21:46

But for this time, it's fantastic for

21:49

static. But yeah, static Netlify is just

21:51

amazing. And just the one where you

21:53

can just drag and drop a project,

21:56

for instance, for static website. So easy.

21:58

But for more, multi-tenant

22:00

dynamic web app like the iJingle

22:03

Maker railway was really cool. And

22:05

as you can see from the

22:07

presentation, it's very user friendly. It

22:11

definitely is. And the pricing is actually really

22:13

decent. They have a free trial and then

22:15

$5 a month. But that

22:17

gets you eight gigs of RAM and

22:19

eight CPUs per service. That is really

22:21

good. I started it on the OB

22:24

plan. And I think that today, just

22:26

to give you an idea, with

22:29

roughly 500 users coming

22:32

to the app per day and

22:34

a lot of them using the

22:36

product, I pay for

22:38

that roughly $15 a

22:40

month for both the

22:43

app deployment and the database,

22:45

which is in MySQL for our iJingle

22:48

Maker. That's pretty impressive. Good

22:51

one to know about this is cool. Yes, nice

22:53

one. They have good Python support. Yes, yes, very

22:55

good. And they have, as I said, like

22:58

most people do today, a

23:00

Discord server monitored also by volunteers. And

23:02

there was a guy the other day, I

23:04

didn't even know he was not part of

23:06

the official team of railway. And it really

23:08

helped me. He wrote some code for me

23:10

to make this thing work with the Celery.

23:13

So it was really, really amazing. And

23:15

it was on a Sunday in

23:18

the evening. So really nice. Yeah, that's

23:20

amazing. One thing I just want to

23:22

give a shout out to, I came

23:24

across recently, that's interesting is Coolify, which

23:26

is it's self hosted

23:28

Heroku Netlify Vercel alternative.

23:30

And you just get pushed to it.

23:33

I don't know if it's better than this, though, because

23:35

you got to have a decent server to run

23:38

it on. You got to run the Docker stuff

23:40

and, and whatnot to sort of do it. But

23:42

yeah, anyway, we get an interesting deployment. Yeah, we're

23:44

going to manage also, because the other day, I

23:46

think it was 10 days ago, there

23:49

was a DDoS attack, I think, again,

23:51

through a way for the European part

23:53

of their network.

23:55

But they were very quick jumping

23:58

on it and managing. all

24:00

of that and doing all the mitigation

24:02

against it. If you run your own

24:04

service, then you've got to take care

24:06

of all of that on your own

24:08

security and all of that. There's benefits,

24:10

but there's also drawbacks there. All

24:12

right, let's talk about, I think the right

24:15

way to start talking about this, how you're

24:17

using AI and the journey you went through.

24:19

I mean, we've talked a lot of the

24:21

foundations already, but let's talk about your app.

24:23

You mentioned a couple of times, AI Jingle

24:26

Maker. Tell people what it does.

24:28

It's a web app that you can find on

24:30

aijinglemaker.com. If you enter

24:32

the app, you can

24:35

create your own sweepers,

24:37

radio jingles, podcast intros,

24:39

audio promos, as if

24:41

you were playing with Lego bricks, I would

24:43

say. It's not

24:45

the usual though experience. If

24:48

you'll be using things like GarageBand or FL

24:51

Studio or Pro Tools

24:54

to edit your audio,

24:56

you can imagine those waves that you've got

24:58

to move around to create

25:01

your jingles or your sweepers. If

25:03

you're using, for instance, the open

25:05

source audacity software also on your

25:07

computer, you can do that. But

25:10

still, you've got to know how to

25:12

move these things around and how to

25:14

play with the volume, the fade in,

25:16

the fade out, the reverb, the chorus,

25:19

the flanger, all the effects. I

25:22

mean, it's not that easy for many people. I

25:25

thought that I could build something

25:27

that made it as easy as

25:29

pressing on a few buttons. That's

25:31

the idea of the product. The

25:35

way you use it when you

25:37

enter the product, you've got four

25:40

different parts. You've got the intro,

25:42

the background, the outro, and the

25:44

voiceover, which is on top of

25:47

the background. The voiceover, now

25:49

you've got three different options. You've

25:51

got the possibility to record your own

25:53

voice. So I rebuilt a voice recorder

25:56

on an online voice recorder where you

25:58

can quickly record your voice. voice, add

26:00

a few effects, and then boom, use

26:02

that as your voiceover. You can also

26:05

upload an existing recording that

26:07

you have on your computer. So if you've

26:09

got something that you recorded with your, with

26:11

FL Studio or something else, you can upload

26:13

it and use it as a voiceover. Or

26:15

you can use AI. So

26:17

I'm connected to a bunch of

26:19

text-to-speech APIs. So I didn't develop

26:22

the models myself. I'm not smart

26:24

enough to develop a foundational model,

26:26

but I'm connected to the APIs.

26:29

You can pick one of the voices.

26:32

You can write your text, generate

26:34

the voiceover. And once you have

26:36

the voiceover, you just select it.

26:38

And when you have the voiceover,

26:41

you choose a background behind it.

26:43

So let's say groovy music, house

26:45

music, rock music, romantic music, whatever.

26:48

You've got some pre-made sounds in

26:50

the app. You choose an intro,

26:52

an outro, and then you press on

26:54

generate jingle. And in five seconds, you've

26:56

got the jingle. That's basically the idea.

26:59

I'll put in the jingle so people know exactly

27:02

what this sounds like. I'll put it

27:04

in right now. It's

27:14

really cool. And it's something

27:16

that I've seen people hire voiceover

27:18

actors on like Fiverr and

27:21

stuff like that. They'll maybe pay 30 bucks

27:23

to have somebody do a couple of trials

27:25

and throw some music on top of

27:28

or something like that. Right. And I'm a voiceover

27:30

artist on Fiverr. I'm doing that also. So

27:33

I know exactly what it is. Let me

27:35

see if I can automate my job away.

27:37

It was not really the idea of automating

27:39

completely a job away. It was more like

27:42

the, how can you create something that self-service

27:44

and that can meet the needs of people,

27:46

as you say, who don't want to spend

27:48

50 bucks or 100 bucks for something, but

27:52

who want to have an easy way

27:54

to create basic sweepers,

27:56

jingles or some other things. And

27:59

the The first target

28:01

was mostly indie radio

28:03

stations or small podcasts.

28:06

But then I saw many brands

28:08

coming, like I wouldn't say, like

28:11

not the big brands, but small

28:13

brands coming like apparel brands or

28:15

coffee shops. Or I had the

28:18

other day a bar in Ibiza

28:21

signing up and paying for credits

28:23

to generate a promo for the

28:25

bar in Ibiza. So

28:28

yeah, it's not only radio stations

28:30

and podcasts. It's also anyone who

28:33

wants to create a

28:35

promo for their business. So that's also

28:37

one of the services. And

28:39

when I started, it was only short

28:41

three-person jingles, radio station IDs, stuff like

28:43

that. But then I added something which

28:46

is called the promo maker, which you

28:48

have in the jingle maker where you

28:50

can create something which is up to

28:52

30 seconds long. So more like a

28:54

radio ad. And

28:57

there you can also use AI. It's connected

28:59

to open AI, to be honest. But you

29:01

can also use AI to generate the text of

29:03

your promo. So just put the name of your

29:05

business, what you're selling, your

29:07

service, your product. Then it will suggest

29:09

the text. Then you choose

29:11

the voice, generate the AI voice, and

29:13

then you add the sounds and then

29:15

it's finished. So let's say in 20

29:17

seconds, you've got a note you add. So

29:19

that's the promise for the promo maker. It's

29:22

a cool app. It's really impressive to look

29:24

at this and think that you built this

29:27

with less than a year of

29:29

Python or really any coding experience, right?

29:32

Roughly that amount. To put things

29:34

really in perspective so that people

29:36

know what they can expect if

29:38

they start their coding journey.

29:41

So I started coding in

29:43

Python in September, October 2022. For

29:48

a few months, as I said, I was mostly

29:50

coding scripts on my computer that

29:53

helped me in my daily business because I

29:55

didn't tell you in the beginning, but most

29:57

of my activities, online marketing, it's a... working

30:00

for startups and bigger companies also.

30:02

So I need scripts, you know,

30:04

for competitive intelligence for SEO, or

30:07

for even content production, this type

30:09

of thing. So I created a

30:11

lot of scripts locally on my

30:13

computer before considering the front

30:15

end part. And that's when I discovered flask,

30:18

I was just before the summer of 2023.

30:20

And yes, the first version of the Jingle

30:26

Maker that I launched, it was just on

30:28

the homepage, you couldn't basically

30:31

tweak anything, there were just four different

30:33

jingles and just put the name of

30:36

your radio station, you press on generate

30:38

and generate it one thing the same

30:40

for everyone. And I started getting some

30:42

traction between May 2023 and

30:46

August. And then I said, Okay, I'm

30:48

going to really dive into this project

30:50

and develop something more sophisticated. I

30:53

started working on that on

30:56

a Monday in September, it was

30:58

released on the Friday, the v2.

31:00

And I gotta tell you, your website now looks

31:04

really nice. You've done a lot of good

31:06

design work on it as well in the

31:08

last couple of months. Yeah, it has evolved

31:10

because maybe you saw it for the first

31:12

time when we decided to do this podcast,

31:15

it was a white page, the

31:17

logo was the same, but it was it was different. Originally,

31:20

it was only in material design,

31:22

I was using material design via

31:25

the CDN from from Google. But

31:28

then I wanted to use Tailwind,

31:30

the new version that you see is

31:32

using Tailwind. So people might recognize the

31:34

Tailwind, Tailwind vibe. And I worked quite

31:36

a lot on this thing that okay,

31:38

we said it's audio, so people don't

31:40

see it. But if you go on

31:42

the website, you will see it there

31:44

is this rotating list

31:47

of things that you can

31:49

create with the with the product. It was inspired

31:51

by some code that I found that I had

31:53

to re tweak it for you know, make it

31:56

responsive and all of that. And

31:58

you out

32:00

to my son with 18, who's learning

32:02

Python a lot. He coded a few

32:05

things recently. He's coding something to

32:07

play darts at the moment, like to calculate

32:09

the scores in darts. And two

32:12

months ago, it said to me

32:14

that your sight is nice, but it

32:16

looks ugly. So could

32:18

you make something that looks better? I

32:21

said, yeah, yeah, but you know, I

32:23

kind of like it. I said, no, no,

32:25

no, no, no, that it's not the

32:28

way it should be. So do something better. I

32:30

know you can do it. So that's why

32:32

also we've got something which looks better

32:35

today. Thanks to my son,

32:37

Max. It's a huge improvement from where it

32:39

started. So nice work on that. Yeah. All

32:41

right, let's talk about some of the techniques

32:43

that you used here to first

32:46

of all, what LLM did you use to

32:48

go on this journey? When I started, I

32:50

wasn't even aware of GitHub co pilots. So

32:52

I started only with the chat

32:54

GPT. I was an

32:57

early premium user of chat GPT.

32:59

So I started paying the 20

33:01

bucks very, very early for for

33:03

GPT plus. So I started using

33:05

chat GPT. So basically switching between

33:08

the chat GPT UI and

33:10

the ID I, I started using a pie

33:12

charm. So that's the first one I use.

33:14

So I was creating some

33:17

code in chat GPT and then jumping

33:19

to pie charm test the code. And

33:21

then if there was a bug coming

33:23

back to chat GPT. So this back

33:25

and forth journey, you ask it at

33:27

the beginning, how do I make a flask app?

33:29

Or what did you ask it in the

33:31

beginning? Like, how do I make a flask

33:33

app? How do I make a flask app

33:36

that does audio? Like what did you I

33:38

knew that I wanted to use flask. So

33:40

the first thing that you would ask is

33:42

to ask chat GPT for a boilerplate of

33:44

a flask web app. And it will give

33:46

you pretty quickly the basic

33:49

building blocks. I don't know if

33:51

it would give you straight away

33:53

the architecture with a template inside

33:56

of the template folder, but no,

33:58

it gives you here. Hello

34:02

world. Yes. But then if you

34:04

know that you need HTML, then

34:06

you will ask a GPT to

34:08

tweak it to render

34:10

index with HTML from from the template

34:13

folder. And that there you see there

34:15

it starts and it gives you some

34:17

boilerplate for the HTML that you put

34:19

in the template folder. And then you

34:22

learn that you

34:24

need ideally to have

34:26

some JavaScript in a JavaScript file

34:28

and that you need some CSS

34:30

in a CSS file and that those

34:32

two files will be in the static

34:35

folder of a Flash Web App. So

34:37

now I've got let's say my routine,

34:39

I've got all of that ready. And

34:41

if I start a new app, I

34:44

don't even need, obviously, a GPT,

34:46

I just create my structure, I

34:48

put index with HTML script.js styles

34:51

of CSS. But then what

34:54

you can also ask GPT

34:56

to do, for instance, is if you've

34:59

got your HTML page, you can

35:01

say restyle this HTML page with

35:03

tailwind CSS using the CDN, for

35:05

instance, I will give you a

35:07

full restyling. And then you can

35:09

ask it, okay, I want on

35:11

this page four buttons, and

35:13

those four buttons should do

35:16

XYZ. And then you start

35:18

moving into the JavaScript part

35:20

of it for the interactions

35:22

on the page. The

35:25

interesting thing maybe, which would explain

35:27

also why I haven't really started

35:30

using next.js or views or things

35:32

like that is that by default,

35:34

for some reason, you've got to

35:37

ask OpenAI why maybe because of

35:39

some legacy importance of a Vania

35:42

JavaScript. But if you

35:44

ask chat GPT to start creating

35:46

some interactions on an HTML page,

35:48

it won't just tell you, okay,

35:50

you know, these days, people use

35:52

react or next.js or something like

35:54

that. So maybe you could or

35:56

you should consider that it

35:58

will go deep into the Vania JavaScript

36:01

and after developing

36:03

the Jingle Mega for a few months,

36:05

I showed it to my friend Alex,

36:07

who is this developer I was talking

36:10

about earlier, was graduated, I would say,

36:12

from Vania JavaScript to React and he

36:14

looked at my code and said, man,

36:17

that's crazy. It's all Vania JavaScript. There's

36:20

not a single line of React. So

36:22

everything in your app is lines and

36:24

lines and lines of JavaScript. That's a

36:26

benefit and a drawback. He

36:29

said that he would not even be

36:31

able to write it this way

36:33

from scratch today. That's not something

36:35

that he would really consider doing

36:37

because it's more verbal. There's more

36:39

stuff to write to achieve what

36:41

you want to achieve. But the

36:43

thing is that with charge EPT

36:45

or today with a GitHub compiler

36:47

embedded in the ID, even if

36:49

it's longer, that's not a problem

36:51

because you don't write a single

36:53

line of code. I think it's

36:55

important to point that out. Is

36:57

that in the Jingle Maker, there

37:00

is not one single line of code

37:02

that I wrote. Not one. That's insane.

37:04

There's no one. Not a single line

37:06

of code that I typed, you know,

37:08

from the first curly bracket to the

37:11

last curly bracket. Nothing. Everything is written

37:13

like that. Yesterday, for instance, I was

37:15

working on something completely different now for

37:17

a friend with also a client who

37:19

were working on a Gmail

37:21

add-on. So I don't know if you're familiar

37:23

with that, but you're using

37:26

Apps Script from Google and

37:28

then create a manifest and

37:30

blah, blah. And I didn't

37:33

know anything about Gmail add-ons until

37:35

yesterday in the morning. And

37:38

after now, five hours of work,

37:40

we've got a fully working product

37:43

in Gmail. That's amazing. That's great.

37:45

Two thousand twenty four. And it's

37:47

only the beginning. I really see

37:49

that evolving super fast. I

37:51

think that now the most important

37:54

thing is to be able to

37:56

talk to the machine, to explain

37:58

the structure, the architecture of. what

38:00

you expect and to be clear in

38:02

your head about the

38:04

code has to be developed for

38:06

the pages to interact for the

38:09

elements to interact together. And if

38:11

it's clear in your head, and if you can explain

38:14

it clearly to the LLM, it's just

38:16

magic. It is like magic. Yeah, it's

38:18

pretty wild. If

38:21

you're a regular listener of the podcast, you

38:23

surely heard about TalkPython's online courses. But have

38:25

you had a chance to try them out?

38:28

No matter the level you're looking for, we have a course

38:30

for you. Our Python for absolute

38:32

beginners is like an introduction to Python plus

38:34

that first year computer science course that you

38:36

never took. Our data driven

38:38

web app courses build a full

38:40

pypi.org clone along with you right

38:42

on the screen. And we

38:45

even have a few courses to dip your toe in with. See

38:47

what we have to offer at training.talkpython.fm or

38:50

just click the link in your podcast player. So

38:54

you can see on the side here, I have

38:56

these really weird recent chats I had like a

38:58

skilled baker Q&A and

39:00

survive a plane crash in the wilderness. And

39:03

so these things are actually related. And what

39:05

I was trying to do was I was

39:07

trying to understand how much prompt

39:09

engineering and sort of setting the stage

39:12

for the questions matters. So like this

39:14

one, I said, you are a very

39:16

skilled baker. You have run a successful

39:18

restaurant in the city for five years.

39:20

Now I will ask you questions. As

39:23

opposed to this one, you are a

39:25

very skilled outdoorsman and you have been on multiple

39:27

reality shows. Now I will ask you some questions.

39:29

And they were both the same question. If I

39:31

am in a plane crash and I need to

39:33

survive, I wanted to see how

39:36

much the baker would focus on food

39:38

and stuff versus the survivalist on others.

39:40

And I feel like that is a

39:43

real clear way to understand this

39:45

kind of stuff in just a real general

39:47

way. But I think there must be a

39:50

code aspect of that as well. Like

39:52

rather than just tell me how to

39:54

do this in Python, like you have

39:56

written this app in Python or you're

39:58

an expert on fast API. Now I

40:00

want to talk to you. You may get different

40:02

results. Uh, maybe better. It may be worse. I

40:04

don't know, but did you try any of this

40:06

techniques or was it just straight down the middle,

40:09

tell me how to do this with tailwinds? To

40:11

be fair with you, because I was a total

40:13

newbie in the beginning, I was making

40:15

sure to ask the AI to explain to me

40:17

almost as if I were, as if I were

40:19

a five year old, you know, because

40:23

I could have told the AI, you're

40:25

an expert, um, Python

40:27

developer and then, uh, just

40:30

give me the code, but then I don't have

40:32

all the explanations because then it

40:34

behaves like an expert Python

40:36

developer, and there are many

40:38

shortcuts that it will take and not

40:40

really give you all the details, uh,

40:42

documenting all the code. So I'm

40:45

still, um, yeah, more like

40:47

in, uh, explain to me in

40:49

layman's terms, you know, how to

40:51

do this, because then also I

40:53

understand the code. So it's

40:55

easier for me to debug it. I usually

40:57

take the analogy that when you said that,

40:59

uh, uh, sometimes you think it might be

41:01

too late to learn to code or to

41:03

learn to do something, let's

41:05

say that you want to learn to play

41:07

music, to play the piano at a certain

41:10

age, it becomes very difficult because you don't

41:12

have the reflex, the muscles agility that you,

41:14

that you can have when you are a

41:16

young kid to, to learn how to play

41:18

the piano. But still you could understand what's

41:20

on the score written on the page from

41:22

the music. So in coding for me, it's

41:25

kind of the same. I would, to be

41:27

honest with you, I would not be able

41:29

to write most of the code that is

41:31

in my app myself, but I

41:33

can read the score and I can

41:35

see the connections between the different building

41:38

blocks in the code. So, uh, I

41:40

can understand if there's a problem, you

41:42

know, a syntax issue because there's some,

41:44

you know, closing brackets that are not

41:46

properly closed. I can see it. I

41:50

wouldn't be able to write the code

41:52

properly in the first place myself. And

41:54

you could argue it's a problem, but

41:56

I say, you know, for me, uh,

41:59

the, the output. is what matters.

42:01

It's the final product which is what

42:03

matters. So as long as I

42:05

understand what's on the screen, for me, it's

42:07

okay. Yeah, well, let's loop it back to

42:09

the original way we started this. One

42:12

way you could do this is you could hire

42:14

people to build this for you, and then they

42:16

give you the code and you're in the same

42:18

position. Like here's a huge pile of code and

42:20

you probably understand it. I'm sure you understand it

42:22

less well than if you went through step by

42:24

step and 100% sure.

42:27

I was in that position a few times

42:29

where I was involved in some projects

42:31

and I was not involved in the

42:33

actual coding of the project. And then

42:35

I inherited the project.

42:37

I had the code

42:40

base, but I could just put it on

42:42

a shelf because I would not

42:44

be able to do anything with that. Whereas now I

42:47

understand every single section of what's

42:49

written in the app. So at

42:51

the moment, I'm repurposing some of

42:53

the payment logic because I would

42:55

switch from Stripe to Paddle in

42:57

the IJingle Maker. And yeah,

43:00

so I've got to rebuild the payment

43:02

models. I've got to rebuild the flash

43:04

crowd for the API endpoints for the

43:06

payment confirmations and all of that. So

43:09

I know where they are and I

43:11

know what they do. So I know

43:13

how to tweak them and I can

43:15

highlight exactly what I want. And then

43:17

now I'm using, I would say 75%

43:20

of the time, I'm using GitHub

43:22

Copilot straight in the IDE directly

43:24

because it's in context then. You

43:26

highlight the code and then if

43:29

the suggestion is nice, just insert

43:31

it to replace the code. So

43:34

yeah, once you've got something that

43:36

you understand, it goes faster and

43:38

faster. Yeah, and you're still early

43:40

in this project, right? This is... Yeah, super early.

43:42

Five, six months or something. Like a baby. Yes.

43:44

I imagine as you do more of this, you

43:46

might not choose to write a lot of code,

43:48

but I bet you'll start to be able to

43:50

write code, kind of like language

43:52

immersion. If you move to

43:55

Brazil, you would start to learn Portuguese.

43:57

Yes. Eventually, just because you're in that

43:59

environment. so much, you know, but sometimes

44:01

really, you know, with the suggestions, it's,

44:06

you wouldn't see why you would write

44:09

the code. For instance, I had

44:11

earlier today, I was creating a

44:13

scraper to check the technology used

44:15

by different websites by looking in

44:17

the source code for let's say,

44:19

WP for WordPress, or do for

44:21

odo Squarespace, we know this kind

44:23

of thing. Yeah, we just wrote

44:25

together with the AI the first

44:28

part, and then we had to replicate the

44:30

code in beautiful sub soup for

44:32

for the scraper for each of

44:34

them. There's absolutely no reason why

44:37

I would write each iteration. And

44:39

once you write two of them,

44:41

then you with co pilot, you

44:43

just go with your tab to

44:46

the next section, and

44:48

we understand that what you expect, because you've

44:50

got like three or four different technologies that

44:52

you want to track in your code that

44:55

you declared before, they will have to write

44:57

a block of six lines. Boom, that's the

44:59

block of six lines. There's no reason in

45:02

the world why I would write those six

45:04

lines. You know, amazing. That's great. That's great.

45:06

One more prompt engineering thing. And you talked

45:08

about and I think it is a problem.

45:10

So we asked

45:12

basically what's the boilerplate and it said, boom,

45:15

here's the boilerplate for flask and chat GPT.

45:17

Then I said, well, how do I use

45:19

a template and it said, boom, here's the

45:21

two pieces of code. You said sometimes this

45:23

can be really challenging. You said one of

45:25

the techniques used was to explain to me

45:27

like a five year old. Another thing that

45:29

works really well is I just said, tell

45:31

me this again, less think step by step. And

45:33

it gave me a 10

45:35

step process. Start one install flask, create

45:38

a project directory, create the project script,

45:40

import flask, and it was really much

45:42

more like this is what each step

45:45

you need to do. And things

45:48

like that are really neat. When you really

45:50

start learning, it's nice to do it to

45:52

do it step by step because you know,

45:54

some people they don't understand. And that's, that's

45:56

quite normal. I would say that you've got

45:58

to install those. libraries

46:00

before using them, that simple things

46:03

like that. So in the beginning,

46:05

it's nice to go step by

46:07

step. It's really powerful. And it's

46:09

something people can check out. There's

46:11

a cool video. That's your channel.

46:13

This one. This one I was

46:15

like, prompt engineering 101 by

46:17

Patrick LaVere. And yeah, he's like a 14

46:19

minute video talks to you, has a bunch

46:22

of advice from places like Anthropic and others.

46:24

And here is how you go and some

46:26

things you can try. It's also evolving. Thanks

46:29

also to the longer context windows

46:31

that we have today in the

46:33

LLMs. Because when I started doing

46:35

that in 2022, the

46:40

context window was maximum 4,000 tokens

46:43

or something like that. So you

46:45

had to really go by short

46:47

section. And it was very difficult

46:49

to debug long pieces of code,

46:51

long sections of code. Now the

46:53

token window is 128,000 on open

46:55

AI. And

46:58

it's even longer on Cloth or other LLMs.

47:00

Today, it's possible to basically

47:03

throw at some point 750

47:05

lines of code in a 1,000 lines of code in

47:07

the LLM and

47:11

get some refactoring, debugging.

47:14

That's also really amazing. When you need,

47:16

let's say you know that you created

47:18

something cool in one app, some business

47:21

logic, something interesting. But you know that

47:23

you want to just repurpose it in

47:25

a small way, like make little tweaks.

47:28

You just throw all the code into

47:30

the LLM. And you just explain the

47:32

little changes that you want. And it

47:34

will give you the adapted code fully

47:36

functional in one go, one shot, which

47:38

is also impressive. That is super impressive.

47:40

So it sounds like you use Google

47:42

Copilot a lot. Have you looked at

47:45

GPT engineer? No. I haven't really either.

47:47

But basically, it's supposed to be kind

47:49

of an iterative thing that

47:51

will generate all the multiple

47:53

files for you and then

47:56

try to understand how all those pieces

47:59

fit together. in work and so more

48:01

like an agent an agent logic. Yeah,

48:04

exactly. Exactly. So but pretty I

48:06

started working with the GPT from

48:08

open AI from from chat GPT,

48:10

I created a few of them.

48:12

It was a funny, funny one,

48:14

for instance, I'm working for an

48:16

accounting firm for some marketing and

48:18

and the boss of the of the

48:20

company sent me a document

48:22

of 350 pages explaining

48:24

the accounting module on module, which is

48:27

an ERP system two weeks ago, and

48:29

he said, Oh, you should read that

48:31

to be able to to understand it

48:33

and answer questions about the module. And

48:35

I just threw it into chat GPT,

48:38

uploaded the PDF and I created

48:40

a custom GPT. So now I

48:42

can add the custom GPT questions

48:44

about the accounting module. So that's

48:46

insane, instead of reading the and

48:49

also another nice use case that

48:51

I have, it's not nothing about

48:53

coding, but sometimes you're just

48:55

on YouTube and someone tells you,

48:57

this is the five big important

48:59

things you have to know in

49:01

2024, which would

49:03

change your life forever. So I'm tempted,

49:06

maybe I will watch this video because

49:08

it might be interesting. But

49:10

you say no, it's probably not that interesting.

49:12

It's probably a lot of fluff. So now

49:15

usually what I do is that I I've

49:17

got a website called I don't remember the

49:19

name, sorry, but one of the websites where

49:21

you can get the transcript

49:24

of a video. So

49:26

I go get the transcript, throw it into

49:28

chat GPT. And and then I asked, what

49:30

are the five important things I should know

49:32

that will change my life forever in 2024.

49:35

And then I've got them in five

49:37

seconds. I don't want to I don't I don't need to watch a

49:39

35 minutes video. And then maybe maybe decide to

49:41

watch it later. Yeah,

49:44

awesome. What an interesting story, Fred. I think

49:47

maybe we should wrap it up. But people

49:49

can check out a i jingle

49:51

maker calm and put that in the show notes.

49:53

There's two other things from you that

49:55

are noteworthy. One is you have

49:58

your AI coding club newsletter. Do you want to about

50:00

that? Yeah, that's a newsletter I started,

50:02

which is connected to some of the

50:05

videos I posted on YouTube, where I

50:07

will explain to people how to use

50:09

AI for coding purposes. I will start

50:11

sending the newsletter soon, actually, I started

50:14

registering people, but I haven't sent the

50:16

first edition yet, but I will start

50:18

very soon. I've got also

50:20

a new app that I set coming

50:22

up, probably within two weeks, it would

50:24

be called the AI Showmaker. It's a

50:27

way to create a podcast or a

50:29

radio show with building blocks. So it's the

50:31

same concept as the Jingle Maker, but for

50:33

longer form. That sounds really interesting. All right,

50:35

cool. People can check it out here. And

50:37

by the way, I've been thinking a lot

50:40

about email and newsletters and stuff. I

50:42

was doing a bunch with MailChimp for

50:44

many, many years, and it's just got

50:46

to be super, super expensive.

50:49

And I got frustrated with them. So

50:51

did a bunch of research on different

50:53

platforms, looked at buttondown.email

50:55

and actually chose a list monk,

50:58

which is self hosted and Docker

51:00

style thing. But one of the ones

51:02

that I came across at the end

51:04

is BeHive, which looks pretty interesting. Are

51:06

you happy with BeHive? It's more competing

51:09

with Substack. I think it was co

51:11

created by one of the guys who

51:13

co founded Morning Brew, the newsletter Morning

51:15

Brew. So he knows what he's talking

51:17

about. And it's really also really well

51:19

coded, really well executed. It's

51:21

a pleasure to use. And

51:24

you can use it for

51:26

free for up to 2500

51:28

subscribers. Yeah, that's

51:30

it. So yeah, the pricing

51:32

is way, way better than

51:34

things like MailChimp and stuff. So also

51:38

cheaper than

51:40

Substack. So they don't, I don't know

51:43

which cut they take from the subscription,

51:46

if you run a premium newsletter, but it's

51:48

much cheaper than a Substack. I think Substack

51:50

takes 10%

51:52

from like Patreon and other platforms

51:54

like that. Nice. All right. And

51:56

then one other thing is your

51:59

call me Fred. Yeah. YouTube channel. Yeah, I've

52:01

got some some videos there. Some of

52:03

videos I explaining how to use the

52:05

agenda maker the recent ones, but also

52:07

some of them are

52:09

about this journey of how

52:11

to code with AI.

52:13

There are some concrete examples also like

52:16

I coded recently,

52:18

a Chrome extension, bookmarking Chrome extension,

52:20

all the code, the explanations is

52:23

on on YouTube and, and the

52:25

code on GitHub. So I've got

52:27

different projects like that. And yeah,

52:30

I'm usually documenting what I do on on

52:32

YouTube. Yeah, nice. So people can check that

52:34

out. All right. Well, I think that about

52:37

does it for the time that

52:39

we got together, but they should come on and

52:41

sharing your story. I think it'll inspire a lot

52:44

of folks. Thank you. I hope that everything was

52:46

clear. I wasn't curious to

52:48

be too cryptic, like the developers I

52:50

was talking about. But yeah,

52:53

so it's, it's really, really exciting.

52:55

It's rejuvenating, as you could

52:57

say, even to open up a whole

52:59

world of possibilities, right? I wish I

53:02

would have started earlier, to be honest,

53:04

because when I see my son, he's

53:06

18. And what he, you know, what

53:08

he already does today with

53:11

Python, that's quite amazing. So when you're

53:13

18, imagine where you will be when

53:15

you will be 48 or 49,

53:19

it will be like really good. So

53:21

yeah, that's the beauty, that's the beauty

53:23

of it. So don't wait too long.

53:25

If you're listening to the

53:27

podcast start today or yesterday, even. Yeah,

53:30

it's amazing. Awesome. Well, totally agree. Thanks

53:32

so much for being here. And

53:34

see you later. Thank you so much. Bye bye.

53:36

Bye. This has been another

53:38

episode of Talk Python to me. Thank

53:41

you to our sponsors. Be sure to check out what

53:43

they're offering. It really helps support the show. Mailtrap,

53:47

an email delivery platform that

53:49

developers love. Try for

53:52

free at mailtrap.io. And

53:55

level up your Python, we have one

53:58

of the largest catalogs of Python video

54:00

courses over at TalkPython. Our content ranges

54:02

from true beginners to deeply advanced topics

54:04

like memory and async. And best of

54:06

all, there's not a subscription in sight.

54:09

Check it out for yourself at training.talkpython.fm.

54:12

Be sure to subscribe to the show, open your favorite

54:14

podcast app, and search for Python. We should be

54:17

right at the top. You can also

54:19

find the iTunes feed at slash iTunes,

54:21

the Google Play feed at slash Play,

54:23

and the Direct RSS feed at

54:25

slash RSS on talkpython.fm. We're

54:27

live streaming most of our recordings these days.

54:29

If you want to be part of the

54:32

show and have your comments featured on the

54:34

air, be sure to subscribe to our YouTube

54:36

channel at TalkPython.fm slash YouTube. This

54:38

is your host, Michael Kennedy. Thanks so much for

54:40

listening. I really appreciate it. Now get out there

54:42

and write some Python code.