Latest / Agentic Conversations (formally mlops.community) / FastAPI for Machine Learning // Sebastián Ramírez // MLOps Coffee Sessions #96

Transcript

FastAPI for Machine Learning // Sebastián Ramírez // MLOps Coffee Sessions #96 · Agentic Conversations (formally mlops.community)

• 0:00
  Give a person, a rose is tamazaki Nal podcast.
• 0:03
  De la comunidad, the envelopes envelopes Community even as a
• 0:07
  toddler and ratico the first EP a-- yatras Casitas, type
• 0:11
  annotations computation. This tree with a possibility at
• 0:15
  this. I'm gonna go ahead and call out
• 0:17
  first and foremost, your shirt. You're wearing the go fast shirt
• 0:22
  for our man fast API. Do some stunts done.
• 0:27
  That's so perfect. Go fast for the go.
• 0:30
  So fast API that's so perfect. I saw that like halfway through
• 0:34
  the conversation and I was like actually realized, but there you
• 0:37
  go, you didn't. Oh man, I thought you did it on
• 0:39
  purpose. My mom.
• 0:41
  Not that clever. Well this was brilliant.
• 0:44
  What were your top takeaways with talking with Sebastian
• 0:47
  Ramirez? I'm a little bit blown away.
• 0:50
  I need to kind of digest for just happened but I think I said
• 0:53
  it right the end that it's very clear.
• 0:55
  That Sebastian has that real deep technical knowledge and
• 0:59
  he's really Honed in a few things he really cares about and
• 1:03
  he can now just bring that value.
• 1:05
  Everywhere we go. Is everything he touches.
• 1:07
  You can add that little sprinkle of magic and its really
• 1:10
  powerful, right? He had some cool ideas around
• 1:13
  like how to tackle those big problems and how to be kind of
• 1:17
  ever mentioned explicitly. But how to stay motivated do
• 1:21
  stuff that you care about, right?
• 1:22
  Might be your own customer. So I really like that he was
• 1:25
  very much like he told us the whole story of how he created.
• 1:29
  Fast API and basically came down to him scratching his own itch,
• 1:33
  right? Even though he tried to not do
• 1:36
  it, he was actively trying not to do it and then he eventually
• 1:39
  does it. And says I got to do it because
• 1:41
  I can't find anything that does what I want.
• 1:43
  And when he rattled off the list of things that he wanted, I
• 1:47
  thought that was brilliant to like, how he knew very clearly.
• 1:50
  Like, I wanted to do this. I wanted to do that, I want all
• 1:53
  this stuff, and I couldn't find anything that worked.
• 1:56
  And then the other thing, like exactly what you were saying
• 1:58
  that really stood out to me the The bigger picture of him
• 2:01
  saying, here's what I'm really passionate about and where can I
• 2:06
  add value with these things that I'm passionate about?
• 2:10
  How can I go into other open source projects now, like what
• 2:13
  he's doing? That will add extra value to
• 2:16
  these projects if I bring my little Touch of Magic to it.
• 2:20
  So that was super cool man. This this guy's incredible.
• 2:24
  Let's jump into the full conversation and for those of
• 2:29
  you that Are just tuning in for the first time.
• 2:32
  We would love it if you give us a follow or a like or wherever
• 2:36
  your listen to this, if it's on Spotify or apple, if you
• 2:40
  subscribe to us and you leave us a little review, it would be a
• 2:43
  huge help. And the other thing that I will
• 2:46
  mention, before we jump into it, is that the apply conference is
• 2:50
  coming up again on, May 18th and 19th.
• 2:53
  I'm going to be hosting it. Surprisingly, they have not
• 2:57
  fired me yet. They I'm The MC and so you can
• 3:02
  expect the Run. Yeah.
• 3:04
  I'm gonna try and do another song before before the end of
• 3:07
  it. And so, there's going to be all
• 3:09
  kinds of good stuff. We've got incredible speakers
• 3:11
  that are coming up at that. Like some of the guests of honor
• 3:16
  that we've had on this podcast. But so many more like really big
• 3:20
  names. Check it all out.
• 3:21
  Will leave a link to the description to register?
• 3:24
  It's free and I'll be there. Accompanying you the whole way.
• 3:28
  Not sure if that's a good or a bad thing.
• 3:29
  NG. But that's what we got.
• 3:31
  Now, let's jump into the conversation that we had, with
• 3:34
  Sebastian remedies, the main thing we should probably start
• 3:41
  with is fast API and Sebastian. For the two people out there,
• 3:45
  listening that do not know what fast API is.
• 3:48
  Can you give us a bit of background on it before we jump
• 3:50
  in? Of course.
• 3:52
  Hey I bet there are many more. That don't know what Sofia is.
• 3:56
  But yeah, so fast EPA is a python framework.
• 4:00
  Building web apis and it's designed for web apis and it's
• 4:04
  built around like all the latest features in the, in the language
• 4:08
  in Python. So, it has a seam support even
• 4:12
  though it's not required to use a sink and it's all based around
• 4:16
  type annotations. So when you define a variable
• 4:20
  and you say, hey, this variable is of this type is string, or is
• 4:23
  a float or something like that. That on one side, that helps the
• 4:26
  editor and other tools, check that your code is correct.
• 4:29
  And it's what allows the editor to give you the completion and
• 4:32
  in errors and all the stuff and then fast EPA uses.
• 4:35
  All those type annotations to a to do data, validation data
• 4:40
  civilization, documentation, and a bunch of other things.
• 4:45
  It's all based on Open Standards for apis.
• 4:48
  And as an end result, you end up having an interactive
• 4:52
  documentation user interface for your API that is based on
• 4:55
  generated automatically and is based on your goals.
• 4:59
  It's from your code. So it's blown up a little bit in
• 5:04
  the last year or so, I would say and you could probably tell us
• 5:08
  when it really started to ramp up.
• 5:10
  But why would you yeah, why do you think that is?
• 5:14
  I like to let you know, like, I never expected it.
• 5:18
  I thought the first EPA will be one of those hipster Frameworks
• 5:24
  that me. And I want your father, 50,
• 5:27
  people around the world use and say like, oh, this is so cool.
• 5:30
  But it's like, no one else that was that was what I was
• 5:33
  expecting and that was actually I guess like one of my highest
• 5:36
  hopes that there will be some other hipsters trying to use it
• 5:41
  with mustaches. Yeah, yes.
• 5:45
  But it seems like the ideas I had on the and it is not just
• 5:49
  ideas. I had is things I learned from a
• 5:52
  bunch of other Frameworks and even other languages.
• 5:55
  And like, I was just trying to put all that together and it
• 5:57
  seems that like those ideas and trying to solve those problems,
• 6:01
  was actually useful and important for many others.
• 6:06
  And I guess that's probably what made it grow so much.
• 6:09
  And I have so much other option that it was something that was
• 6:13
  useful and It's kind of necessary for a lot of people.
• 6:17
  Yeah, no, I'm not. I think, like, when I first
• 6:19
  started hearing about, because the fault was flask, right?
• 6:23
  Everyone was just like, right, you learning data science, where
• 6:25
  I came from. I said, was, I right learned
• 6:27
  eight signs right? Need to be like Pfizer, you got
• 6:29
  a bun API. They are this flossie tutorial
• 6:31
  blah blah blah where was The Tipping Point?
• 6:35
  Because all of a sudden like all the Articles now is like right?
• 6:37
  It's fast API. That's the go-to and things that
• 6:40
  I did. You feel that happen or did it
• 6:42
  kind of happen gradually. It's like, I feel like it's been
• 6:47
  happening gradually and like, just, like going happening, more
• 6:50
  and more. But, you know, like, like I came
• 6:54
  from from flask. The best thing I had before, I
• 6:57
  had built fast API, was flask with a bunch of plugins, it's
• 7:00
  just that it was a very fragile, it's set up with, you know, like
• 7:04
  trying to have plugins interact with each other.
• 7:06
  That are not made to be working with each other and like, its it
• 7:10
  was quite more cumbersome but it's like From all that work and
• 7:16
  from all those plugins and all that open source contributions
• 7:20
  from other people, is that I learned a lot of things that
• 7:22
  they like was able to apply to fast API but actually, I don't
• 7:26
  know. I feel like it's been just like
• 7:28
  steadily growing and steadily happening that is just like
• 7:33
  becoming these things that is they go to in many cases for
• 7:37
  these types of apis. But they're like, I don't know,
• 7:41
  it's it's still mind-blowing. For me to see how many people
• 7:47
  are using it. And for how many things like
• 7:49
  from time to time, I hear like this.
• 7:51
  I don't know. Some huge corporation and
• 7:53
  someone says, like, oh, we used Fast API, like I had no idea.
• 7:58
  This is very fun. Yeah.
• 8:00
  You're like, are you sure about that?
• 8:03
  Yeah. Are you sure you want to do you
• 8:08
  trust me? That's awesome.
• 8:11
  So, there is a One thing that I want to comment on with that is
• 8:16
  or ask the question about it's around what has been some of the
• 8:20
  most surprising ways or use cases or people like you said
• 8:25
  corporations that are using flask.
• 8:27
  All right. Yeah.
• 8:28
  Fast API. Sorry not fussed.
• 8:31
  Well, flask. I know there's a lot of people
• 8:33
  using pleasant but they I don't know for fast APA.
• 8:37
  Think that there's, there's several things that have been,
• 8:39
  like, completely unexpected. For example, I know that the
• 8:44
  They in it telescope Institute in the United States where they
• 8:48
  like use that to process the data of the telescope's of the
• 8:51
  world and all that stuff and they are using fast API or, I
• 8:55
  don't know. Netflix Disney over Microsoft
• 8:58
  that there is in my for Windows apis or at least, at least
• 9:03
  that's what I what I have heard like.
• 9:06
  What someone told me once, I don't have access to those
• 9:12
  companies directly to many of those Those to know what they
• 9:15
  are using now, how they are using it now but you know, like
• 9:18
  there's a still, like a bunch of, I don't know, big, big,
• 9:20
  pharmaceutical companies, things things like that.
• 9:25
  The yeah, I guess I never expected it to, to grow so much.
• 9:29
  And some of the strangest use cases.
• 9:32
  I have seen are, for example, streaming.
• 9:35
  I remember someone who wants made a project that was
• 9:38
  streaming the terminal, like, some some common thing terminal
• 9:43
  through. Web browser using websockets and
• 9:47
  using fast API to handle all the stuff and to handle the
• 9:49
  communication between the terminal on the browser and it
• 9:51
  was like wow this is this is a crazy and so cool.
• 9:54
  Like I didn't imagine you could do that.
• 9:58
  Yeah, I don't know or also to transfer the data to simulate
• 10:01
  galaxies. I remember some people were
• 10:03
  doing something like that. I don't even know what that
• 10:05
  means, but it's very cool. That's not.
• 10:10
  So I guess that's always going to happen when something gets
• 10:12
  public domain and then Everyone say you zoom and twist it and
• 10:15
  break. It has like the opposite
• 10:19
  happened as well. Have through its popularity, I
• 10:23
  guess, you've got more people wanting to contribute wanting to
• 10:26
  do stuff and change staff house, that kind of has anyone done
• 10:31
  anything this blown your mind or you've looked at and you've gone
• 10:34
  all? Wow, like I never thought of
• 10:35
  doing that, that way. And what's the biggest sort of
• 10:38
  lesson learned and things from that experience?
• 10:41
  Huh. Hey, there's been a lot of
• 10:43
  things that I have ended up like, having to learn a lot, and
• 10:46
  having to understand a lot of the deep Concepts and ideas, but
• 10:51
  like, you know, like one of the biggest of those was actually,
• 10:54
  while I was still building faster PA seeing the code inside
• 10:59
  of pedantic, but I don't think is so powerful and so simple to
• 11:03
  use, it's such a good library. And, you know, it's a great like
• 11:06
  very outside the first APA on its own is just amazing.
• 11:09
  I think. All of the data Science World
• 11:13
  could benefit from pedantic to be able to have like automatic
• 11:16
  data, validation, and type safety, and all the stuff and in
• 11:19
  many other use cases. But so for the, for the, for the
• 11:23
  developer, that is using pedantic as a user, it's
• 11:26
  amazing. But the internal code of
• 11:28
  pedantic has to do. So many, complex logic, and so
• 11:33
  many complex tricks to be able to do is thing that I remember
• 11:38
  when I was working on pedantic itself out that like trying to
• 11:40
  control Tribute something like that.
• 11:43
  It was so mind-bending, like, all the things that it was doing
• 11:45
  and how it was interacting with the internal parts of how python
• 11:51
  works with data and with with objects and all the stuff that
• 11:54
  that was that was I think that was one of the most challenging
• 11:58
  Parts in the like the whole process.
• 12:01
  How how does it take to like, how long does that take you?
• 12:05
  Right to learn enough to be able to comprehend that?
• 12:09
  Because I you talk even that's get Get nervous you talking
• 12:12
  about it thinking God like I've seen big complicated tools and
• 12:16
  dive into the code base and it can be really intimidating to
• 12:18
  get even started with like what was that?
• 12:20
  How long a process is that? Or is it something you chip away
• 12:23
  at? You can add to that.
• 12:25
  Look at an oil-less. Like the things that the process
• 12:29
  with fast API in general was kind of strange because it's not
• 12:33
  that I said, one day like, oh, I'm gonna build faster pi, and
• 12:36
  then I just started building the thing.
• 12:39
  It was actually like, you know, years, Years of using many
• 12:42
  different Frameworks. I try to use like almost all of
• 12:45
  them. I was avoiding building
• 12:46
  something new for a very long way and I was like, no, no, no I
• 12:49
  cannot I cannot try to let's not make this another ecosystem
• 12:55
  where you have a new framework every every two weeks.
• 12:57
  So I was trying to avoid building something new and
• 12:59
  trying to just use the things that were available in trying to
• 13:01
  find the best like The Sweet Spot of all the features and all
• 13:05
  the things that I wanted to have.
• 13:08
  And then when I found the right tool for that, The outer had to
• 13:11
  deprecated and that was like, I think that that was the last the
• 13:15
  last drop to say, like, okay, let's try to build this thing,
• 13:18
  but then there was a lot of time, just stalling the
• 13:21
  standards. So, first API is based on open
• 13:23
  API, which is a standard for building apis and that in turn
• 13:27
  uses another standard called Json, schema.
• 13:30
  And it also uses it, which is for defining Json, payloads for
• 13:33
  defining the the content of Json objects and that that open API
• 13:40
  standard also Uses the other standard called oauth 2 which is
• 13:43
  for doing authentication and authorization.
• 13:45
  So it was just a long time, just studying those standards to try
• 13:49
  to understand. Okay, how can I make something
• 13:52
  that is built around this? And that is base uses these
• 13:54
  Concepts and like, reuses them. Unlike exposes them in code in
• 13:58
  the simplest way possible, and it was just like a lot of
• 14:02
  learning and just a lot of studying and trying to figure
• 14:05
  out what was the way best way to that.
• 14:07
  Now, that then after, after all the process of learning, I think
• 14:11
  the the standards he was actually just well, I decided,
• 14:14
  let's go with Starlit and style. It was made by the same person
• 14:18
  by the same time Christie which is amazing.
• 14:20
  The same author of this previous library that I thought was the
• 14:24
  perfect thing. The private library was called a
• 14:26
  pi star and he had to replicate it and don't Christie.
• 14:30
  The a was creating Starlet which was this very high performance
• 14:34
  micro framework. / toolkit for python web development.
• 14:38
  He is the same author of Django rest.
• 14:41
  Framework. So is a lot of a lot of
• 14:43
  learnings from building apis starting from the jungle world,
• 14:47
  taking that over to a pi star, which was this previous
• 14:51
  framework. And then like just focusing on
• 14:53
  the micro framework and then I was just taking all those
• 14:56
  learnings and, like, in everything all that all that
• 14:58
  work to build fast API on top. But then, I wanted to mix it
• 15:02
  with python, take that had all these great user experience on
• 15:06
  like interface to work with data, just using standard type
• 15:10
  on attention. So that Don't have to import and
• 15:12
  learn some new class for defining strings.
• 15:15
  You just use the standard python string, you could use the
• 15:17
  standard python float, and all those things.
• 15:20
  And then it was just also a lot of work in pedantic implementing
• 15:24
  Json schema which is the solar standard in my tactics
• 15:27
  orthodontic will have support for Json schema so I do the
• 15:31
  stuff to the NBA then finally be able to integrate that Json,
• 15:35
  schema from pedantic, aim to open API and put all those
• 15:38
  things together. So let's just unload Lot of
• 15:41
  building the the recipe ingredients and like making sure
• 15:44
  they were in place before actually cooking the thing.
• 15:48
  So it was like, I don't know, like the moment when I started
• 15:52
  building fast API itself, I have been working on trying to make
• 15:56
  that thing happen for quite a while.
• 16:00
  And what were the main pieces that you felt like we're missing
• 16:04
  when you were out there window shopping?
• 16:07
  What was it? That was like, damn, this has
• 16:10
  this. This but then it doesn't have
• 16:11
  this. And then maybe there's another
• 16:13
  one that does have that and then it doesn't have this.
• 16:15
  And so you need to have them both but you can't use them both
• 16:18
  and what were those pieces that you were like I really need
• 16:22
  this. So that there's actually a bunch
• 16:28
  I wanted to have, I wanted to have to start.
• 16:30
  I wanted to have auto-completion.
• 16:32
  Hey, an in Liners, in the editor, I have really bad memory
• 16:36
  and I want the editor to tell me, what is the thing that I
• 16:39
  have to type? I know, when I, when I'm trying
• 16:42
  to make string is in uppercase letters, I never remember it.
• 16:45
  See topper, is it? Offer case is it capitals?
• 16:48
  I don't remember the names of the methods even more when you
• 16:52
  are working with. Python that like one of the best
• 16:55
  features of python is that you have the flexibility of using so
• 16:58
  many libraries and so many packages to for so many things.
• 17:01
  So you're normally just using new packages and new libraries.
• 17:04
  There's no way to just memorize all the methods in all the
• 17:07
  things that you are using. So I really like being able to
• 17:10
  have a to completion, I wanted to have that.
• 17:12
  I wanted to avoid as much as possible code duplication and
• 17:17
  information duplication you know, in in many Frameworks and
• 17:20
  in many tools you have to say in one place.
• 17:22
  Hey, Name is a string and age is a float, but then again you have
• 17:27
  to say the same information in another place for in one for
• 17:31
  documentation. In another one for data,
• 17:33
  validation in, another one for serialization, for the output of
• 17:36
  the data and maybe in another one to save the thing in the
• 17:38
  database. And you have to use four
• 17:41
  different ways of declaring that something is a strength, which
• 17:45
  is not necessarily code duplication because you are
• 17:47
  doing it in a, in a different way, using a different class,
• 17:49
  from whatever specific libraries, but this is still the
• 17:52
  same Meta information, sort of, but sir metadata of that
• 17:56
  particular model of that particular data type.
• 18:01
  And I wanted to avoid all that duplication of information
• 18:04
  because keeping that instinct is very, very error-prone and it
• 18:11
  introduces a lot of boats in the future when you have to refactor
• 18:14
  and change things. So I wanted to avoid that, I
• 18:16
  wanted to have a best practices by default, I wanted to be able
• 18:19
  to have data validation and to let you know when I was building
• 18:23
  Simple apis, with flask. And with other tools, the twins
• 18:28
  are getting started guide in those.
• 18:30
  In those in those tools into spring, a is pretty
• 18:33
  straightforward, but then you quickly realize that you have to
• 18:36
  validate that the data is correct and that what you're
• 18:39
  receiving is actually correct. And then you end up having 50
• 18:42
  lines of code with if statements everywhere trying to check if
• 18:46
  this is actually a string or not.
• 18:49
  And I wanted to be able to have that solved by default.
• 18:53
  A I wanted to, I wanted so many things I wanted to have very
• 18:57
  simple code. I wanted it to be very short so
• 18:59
  that you didn't have to create an instance of a class to them
• 19:02
  pass. Another instance, to another
• 19:03
  function to some, you know, I wanted to have great support for
• 19:07
  all the electors. So I and that was actually part
• 19:11
  of the process of building fast API before writing the internal
• 19:14
  code, I wrote the design of the interface that the final user
• 19:19
  will actually use for and I tested in several Laters, I
• 19:23
  tested with Visual Studio code with pycharm, and with editors
• 19:26
  based on Jedi, which are like most of the others to make sure
• 19:29
  that like everyone will have the best developer experience
• 19:32
  because that's what I was targeting.
• 19:34
  Like I was targeting myself as a developer.
• 19:36
  I wanted to have a good time but then I wanted to take if anyone
• 19:39
  else is coming from any other later, I want them to have a
• 19:42
  good time developing with this thing as well.
• 19:45
  And I wanted that, I wanted to have a automatic interactive API
• 19:50
  documentation. I thought that was amazing and
• 19:52
  it's the That was the first time I saw that was with Django rest
• 19:54
  framework. And then I realized, then I
• 19:57
  discovered that there was a new a new standard for defining apis
• 20:01
  with open API. And it has swagger UI that cool
• 20:04
  interpret that definition of the API and provide this amazing
• 20:09
  interface. So I wanted to have the thing
• 20:11
  based on that and I wanted the wanted to be able to have good
• 20:15
  security tools and authentication.
• 20:17
  Authorization, roles permissions, whatever was
• 20:19
  needed. I also wanted it to be as
• 20:22
  independent. Pendant as possible from any
• 20:24
  other tools out. So it wouldn't need plugins.
• 20:28
  It wouldn't need any particular database or any particular like
• 20:31
  you can just like import the package that you want to use and
• 20:33
  just use it. Direct.
• 20:36
  I want, I don't know what else I wanted to.
• 20:37
  I wanted to have dependency injection to and I actually I
• 20:40
  brought that from the front and World working with typescript
• 20:44
  and I wanted to have this, this this features here as well.
• 20:48
  I think there was a bunch of other things but I'm already
• 20:51
  talking too much. So, yeah, a little It's funny
• 20:54
  because we kind of going through some of this been through this
• 20:58
  kind of discussions around language Choice, until choice in
• 21:01
  the past in places of worms and it's almost like you wanted it
• 21:05
  to just not be poison almost 4 - like that's all the stuff that
• 21:09
  you kind of get from by some of them on the turtles and other
• 21:12
  Frameworks and stuff like that. Or flip that on its head, you
• 21:15
  wanted to make a really mature tool but for python, right?
• 21:19
  Which is a really, really cool like Nobles of endeavor, Over.
• 21:23
  And that kind of leads me down the path of thinking as a, like
• 21:27
  a standard Dev, who's maybe quite experienced, but
• 21:33
  frustrated with the tools right? And they're, they're going to
• 21:36
  build the next fast Epi, right? But they don't know where to get
• 21:39
  started. What kind of advice would you
• 21:41
  give like to? Yeah.
• 21:42
  Someone that's technically capable probably has the time
• 21:45
  and inclination, but just doesn't know what to get going
• 21:48
  with. Is it just stopped?
• 21:49
  Building, is it contribute to other people's stuff?
• 21:51
  Is it find a community? The where would you go or what?
• 21:54
  Advice would you give? So the, well first, I don't
• 21:58
  think I'm great at giving us back, but I think that they main
• 22:03
  advice, I normally given to everything is to focus on
• 22:06
  solving a problem if you can find and, you know, like, I have
• 22:12
  this may be kind of controversial opinion that
• 22:15
  there's a lot of like, I don't know, a lot of information and
• 22:19
  study and things are run innovation.
• 22:22
  And disruption and things like that.
• 22:25
  And I don't like those words very much because I feel those
• 22:28
  are side effects of solving a problem I consider the main
• 22:33
  thing is actually solving a problem.
• 22:34
  That's a great Point. Yeah and the that also helps me
• 22:39
  focus a lot more on what to do, you know?
• 22:42
  Like when you're trying to learn something it's like okay like
• 22:45
  maybe I should be a master in python or something like that
• 22:48
  and it's that's a painful path because you can never learn
• 22:51
  everything there is to know Do you know about the python or
• 22:55
  almost any other area is in particular, if you in
• 22:58
  technology, but then being able to focus on a particular problem
• 23:02
  that is important to you in Nikes like something that was
• 23:04
  important to me that I wanted to solve that help me a guide.
• 23:09
  Like should I learn this or not? Is this actually really going to
• 23:12
  contribute to solving the problem I'm interested in or
• 23:16
  not? And that can help a lot like
• 23:18
  decide what to do and what not to do on the other hand on
• 23:22
  there. And on the other side, if this
• 23:25
  problem is important to you and it's important to you as a
• 23:28
  developer. In this case, there's a high
• 23:31
  chance that it will be important to older people as well.
• 23:34
  And if you find a way to solve it in a nice way, it's probably
• 23:39
  going to be helpful to others and as part of solving a problem
• 23:45
  than the first thing that I try to do on the first thing that I
• 23:48
  think is useful to do is try to find existing Solutions.
• 23:51
  So be becoming a Power user before trying to build something
• 23:55
  from scratch because there are many solutions that are already
• 23:57
  great. And that can kind of really
• 23:59
  cover a lot of use cases. And a lot of things that maybe
• 24:04
  are not obvious to you or to me like when building something
• 24:07
  from scratch. So trying to learn like how to
• 24:10
  use the things that are rare and that my daughter is of the
• 24:13
  problems. I think that's a great start but
• 24:15
  then if you see that, it's not really solving.
• 24:18
  The problem on it is not really solving it in a way that is
• 24:21
  convenient enough for his notice.
• 24:22
  As close as the thing that you wanted to have enough, then then
• 24:26
  it probably makes sense to just try to build something that sort
• 24:30
  of satisfies that but focusing on solving that problem.
• 24:34
  Yeah, I loved it. You scratch your own itch but
• 24:37
  before you scratch your own itch, make sure that there's not
• 24:40
  a back scratcher there and like absolutely that.
• 24:43
  I think that's a Perfect Analogy.
• 24:46
  So it's and that makes a lot of sense on what you did, your path
• 24:50
  was I'm going to go and try and do something like.
• 24:54
  I really don't want to build anything that was your
• 24:56
  mentality. It's like again I really would.
• 25:00
  If I can I'm going to stay away from trying to build something.
• 25:03
  And then you went and explored and explored and explored until
• 25:06
  you got to the point where you thought, oh well I can use this
• 25:10
  and lo and behold that what you wanted to use was getting
• 25:14
  deprecated in you're like shoot man, maybe I need to build
• 25:17
  something, maybe has totally so I'm wondering.
• 25:21
  How do you go about making Fast API better?
• 25:25
  Are you doing? Like I can see a world where
• 25:28
  you're spending your 9 to 5 in like meetings or user meetings
• 25:34
  user research trying to figure out what the next best thing is
• 25:38
  or is it still just you're trying to scratch your own itch
• 25:40
  every step of the way. This is the low bunch of things
• 25:45
  that I already have in mind under.
• 25:48
  I already want to improve under your body is see that that that
• 25:52
  I want to happen in particular, to simplify the developer
• 25:58
  experience to simplify things that are already that can be
• 26:02
  improved. And I think like a lot of that
• 26:06
  is already like on the, on the, at least on the short-term part
• 26:10
  back. Look.
• 26:10
  But then, at the same time, I'm also.
• 26:12
  So here in a lot of people asking for things and like, you
• 26:16
  know, for example, right now, I'm in the middle of this
• 26:19
  complex, refactor to improve some things to do to allow doing
• 26:24
  certain things with routers that with some particular internal
• 26:27
  component for those that don't know much about fast API.
• 26:31
  So some internal part that you would normally use in one way.
• 26:33
  I want to extend that to allow using it in more ways.
• 26:38
  But then like you know this this reflector touches so many
• 26:42
  Things. And so many.
• 26:43
  So many particular Corner cases that have come along in previous
• 26:48
  issues and previous pull requests, and that already have
• 26:51
  tests there and things like that.
• 26:54
  In many cases thing things that I'm not personally interested
• 26:58
  in, but I know that there are some other uses that are
• 27:01
  interested in having those features.
• 27:03
  So, it becomes increasingly more complex to make sure that I
• 27:08
  satisfy like all the needs of everyone who they are.
• 27:12
  It's, it's an interesting Challenge and it's an
• 27:15
  interesting problem and at the same time there's also a bunch
• 27:18
  of things that like I have been getting more and more involved
• 27:24
  in the type annotations with python and these things.
• 27:28
  And I get to see and to understand a bit more every time
• 27:32
  a little bit more, how all these things work and how I can
• 27:35
  improve, all the type annotations and how they are
• 27:38
  used internally in fast API, how to make things easier and more.
• 27:42
  More obvious and more intuitive, you know, all these things.
• 27:46
  So, so, yeah, there's a bunch of things that I want to improve in
• 27:49
  that to make like all the experience and all the
• 27:52
  development nicer, What specifically are the things that
• 27:56
  you want to simplify and and why are they cumbersome right now?
• 28:00
  So for example, there's some particular parameters that you
• 28:04
  can pass to for example, to all those objects and they receive a
• 28:09
  dictionary that will have some particular variables.
• 28:12
  And some particular. So particular keys with some
• 28:15
  particular types and right now in the internal Force FEI, go.
• 28:19
  This, this is defined as a python Tech model, but you can
• 28:22
  also pass a dictionary with the same fields of the pedantic
• 28:25
  model. But the thing is, if you start
• 28:27
  typing the dictionary, you will not have Auto completion for the
• 28:31
  kids that you will have their. And now there is a there is a
• 28:34
  new world relatively new type annotation in Python.
• 28:37
  That is called typed dict that allows you to define something
• 28:42
  like a pirate. Mobile support for refining
• 28:44
  dictionaries, and the keys, and the types of the values of those
• 28:47
  keys for dictionaries. And if I can Define this
• 28:51
  parameters, as type, the eggs, then tools that can check for in
• 28:58
  liner for, for type errors can show in line errors when you are
• 29:01
  passing an incorrect key and things like that.
• 29:03
  And recently like I have been so excited about this type things
• 29:07
  because I knew that this will allow editors to provide better
• 29:11
  auto-completion. Or dictionaries Visual Studio
• 29:14
  code, had support for that for a while.
• 29:17
  And now pycharm recently added support for auto completion and
• 29:20
  Inland areas for these tactics. So now you can have you can
• 29:23
  start a dictionary for example. Let's say you're calling a
• 29:25
  function and then you say, you're going to pass the
• 29:28
  parameter and this parameter, let's say is called options and
• 29:31
  options equals. And then you start the
• 29:33
  dictionary and then you hit command space control space or
• 29:37
  like, whatever is the key to trigger to completion.
• 29:39
  And then the editor now in in modern, There's and now it's a
• 29:44
  python added that like last week or two weeks ago, this very
• 29:47
  reason. Now they will provide you how to
• 29:49
  completion for the keys of the dictionary that you can pass to
• 29:53
  that dictionary options and things like that.
• 29:56
  Of course, it requires that the options parameter is annotated
• 30:01
  as that particular type dict. So, I want to add type tickets
• 30:05
  to a bunch of the parts in fast API that will receive options,
• 30:09
  so that so that you can have more of the conversation.
• 30:11
  I'm always optimistic. My signal to confusion, I guess.
• 30:14
  I love that. So I've got like a more
• 30:17
  envelopes. Focus question that I really
• 30:19
  want to ask, but part of me, really wants to ask you've
• 30:23
  talked a lot about scratching your own agent solving problems
• 30:26
  that frustrate you, how what does it do for a festival on?
• 30:30
  How much of your time do you now, spend building part fast
• 30:35
  API versus using it to actually solve the problem?
• 30:39
  You know, you know, that that's also, that's also, that's also
• 30:42
  something that I find. Very, very interesting
• 30:45
  everything. Faking is that many open-source
• 30:47
  developers, end up spending much more time, maintaining the open
• 30:51
  source code than working with it.
• 30:54
  And I feel that I feel that is, I feel that is dangerous because
• 31:00
  implicitly, you will end up trying to optimize reducing the
• 31:04
  the work and like scratching your own each as maintainer of
• 31:08
  the code not as I user of the code.
• 31:10
  So I try to make an effort to do things that require me to use
• 31:15
  Stacy I as a developer so that I can, I can feel the pains that
• 31:20
  developers will have and that I can figure out like, okay, I
• 31:23
  need to improve this thing. I need to improve this other
• 31:25
  thing, for example, another of those things that I want to
• 31:27
  improve is that when you declare pedantic model, for example, for
• 31:31
  a response and the data that you are returning back from the
• 31:35
  function is not valid. Then that means that the data is
• 31:40
  a server error, because you are saying you are telling the
• 31:42
  client. Hey, I'm gonna give you this
• 31:43
  information but then you are giving it all the So the error
• 31:47
  says like, hey, this is invalid. I'm not going to just let it
• 31:50
  pass through and send the crying something that could be invalid.
• 31:54
  And like that is breaking the contract or something like that.
• 31:57
  That server error, currently shows a very long stack Trace.
• 32:02
  That doesn't give any of information of where the problem
• 32:04
  exactly. Is that something that I want to
• 32:06
  improve and that's something that I get to see when I'm
• 32:09
  working with a big Festival application and it's like, oh
• 32:12
  no. Where is this problem?
• 32:14
  Where is this happening? And then like I have to go
• 32:16
  through the pain of finding out where the thing is.
• 32:20
  But seeing those things as a user from the perspective of the
• 32:24
  user, that is using the tool helps a lot.
• 32:27
  Hey helps a lot in guiding again in also guiding the direction of
• 32:32
  like, where to go next and what to improve on what to do.
• 32:36
  That must be really light. Must be a really useful sort of
• 32:38
  experience, but at the same time, you can't really be angry
• 32:41
  at anyone because you put it there.
• 32:44
  So, right way to get onto the develops question then thinking
• 32:49
  about it. So if it wasn't fast API, right?
• 32:51
  So say that already existed someone else was doing here.
• 32:54
  Yep. Do you have a feel for what the
• 32:56
  what would be the next each you were going to scratch in like
• 32:59
  them? A lot of space?
• 32:59
  What do things missing? What?
• 33:01
  What frustrates you in that space?
• 33:03
  That you think? Yeah, actually, I could take a
• 33:05
  good stab at this. Build something that that would
• 33:07
  complete. That's all set, you know like in
• 33:11
  general in the envelopes area or like in apis.
• 33:16
  In just in general, in the envelope series but I think both
• 33:18
  would be interested in whatever. Yeah, you know, you know, and
• 33:21
  this is actually quite quite fun.
• 33:24
  I just starting working a little bit with, with any scale, the
• 33:28
  creators of Ray, which is this distributed computing and
• 33:31
  distributed machine learning and think General distributed
• 33:34
  computing system for Python. And I think that's that's an
• 33:37
  area that can that can benefit from like, you know, like these
• 33:41
  are the completion and type errors and like all these
• 33:43
  things, all this, all this All these things that I'm always
• 33:48
  obsessed with and they actually doing a great job with that but
• 33:52
  I still like, you know, like being able to see if there's
• 33:55
  something that can be improved and like I'm trying to do over
• 33:58
  there. It's amazing and that's
• 33:59
  precisely what I'm what I'm going to do with them just like
• 34:02
  check the communication and check the the like if there's
• 34:05
  anything that can be improved in the type annotations so that the
• 34:08
  developing distributed applications can be as nice and
• 34:13
  enjoyable and safe as possible. Safe in terms of like being sure
• 34:18
  that your code is not broken and that you're not using incorrect
• 34:21
  types and that type of thing. So, yeah, that that's something
• 34:25
  that I think is very, very cool. Where do you feel like right
• 34:32
  now? The landscape is going, how do
• 34:37
  you see it changing over the next one year?
• 34:40
  Five years, one thing that I feel is gonna is gonna is gonna
• 34:48
  improve a lot and it's going to be great.
• 34:50
  Is that recently? There was a it was a accepted a
• 34:53
  new way to defined type annotations for n-dimensional.
• 34:57
  Erase or and erased answers. However, you want to call them
• 35:02
  and to Define as part of the typing information to define the
• 35:07
  sizes of these tensors of these Cynthia Racers, please whatever
• 35:13
  that means that at some point in the future.
• 35:17
  Because like, that needs to be accepted as part of the language
• 35:20
  of as part of python itself before it can even like be used
• 35:23
  by libraries. But now that it was accepted,
• 35:26
  I'm pretty sure it was accepted me and And things of the ground,
• 35:28
  but I'm pretty sure it was Exempted.
• 35:30
  Now, a libraries can actually use it.
• 35:33
  And that means that known by tensorflow by torch, all these
• 35:36
  things can start, adding type annotations with that
• 35:39
  information and that will mean that when you are at a know,
• 35:43
  when you are multiplying 1, Matrix with another in Mumbai or
• 35:48
  tensorflow or whatever, you will be able to have any nerves right
• 35:52
  in the editor. Later, red squiggly line telling
• 35:54
  you, hey, there's this size mismatch in this.
• 35:57
  Two matrices, they are not compatible.
• 35:59
  You cannot multiply one by the other things like that.
• 36:02
  I think that's that's going to be amazing.
• 36:04
  Like, you know, because right now, a lot of that stuff can
• 36:07
  only happen at runtime, you have to run the script to then see
• 36:10
  how he explodes and then try to figure out, where is the
• 36:14
  problem. And like, for example, that
• 36:16
  likes a mismatch of sizes is something that is normally very
• 36:21
  difficult to the book. Like, it takes a long time to
• 36:24
  realize like, oh, of course, I have to flip the The this this
• 36:29
  Matrix first because I'm multiplying it incorrectly or
• 36:32
  like I put it in the wrong side. I have to multiply the the
• 36:36
  other, by this one, not this one, by the other or things like
• 36:38
  that. You know, like that in many
• 36:40
  cases are just like simple do mistakes, but being able to
• 36:43
  notice that very quickly, right? In the editor of writing the
• 36:46
  point where that is happening, I think that's going to be like a
• 36:49
  massive productivity booster. I think that's going to be,
• 36:53
  that's going to be amazing. The other thing that I fail is
• 36:56
  probably gonna happen, and And that I hope can happen is a bit
• 37:01
  more of interaction between the machine learning world and the
• 37:04
  async python world. Because right now, most of the
• 37:08
  machine learning and ml mL of some data science and all these
• 37:13
  types of things, apis and packages are blocking and you
• 37:18
  never use. AC generate, you just like, Cody
• 37:20
  thing directly, but been able to mix, both and being able to and
• 37:24
  learning how to do that properly and how to use it.
• 37:28
  Going to wait and at the same time, use much-enduring, that
• 37:30
  will allow a lot of improvements in performance and in terms of
• 37:35
  concurrency and in terms of distributed applications and
• 37:38
  things like that. So I think that's something that
• 37:40
  cool could improve a lot and python is made to like the way
• 37:47
  python is growing right now makes it such a good language
• 37:51
  for doing concurrent code for doing a sink and a weight
• 37:55
  things. And for interacting with that,
• 37:58
  And that plus the fact that python is the language for
• 38:01
  machine learning. I think it's going to be a great
• 38:04
  combination so then what things aren't going to change.
• 38:09
  Like if you had to bet, this will still be the same in five
• 38:13
  years and not because you want it to just because you feel like
• 38:16
  this is what we've decided on. Maybe what would you say?
• 38:20
  Where would you put your chips? What would not change?
• 38:30
  So for example, I know that there's most certainly not going
• 38:37
  to be a breaking change, like the change from python 2.2,
• 38:41
  whatever to Python 3 point whatever because that the, that
• 38:46
  made a lot of code in incompatible, with whatever was
• 38:50
  the previous versions and that the was a very painful
• 38:55
  transition and it Very painful for the maintainers of python
• 38:59
  itself. Like, you know, having to deal
• 39:00
  with all that having to deal with all the users and all the
• 39:04
  things like that was, that was something that was a very
• 39:08
  painful transition that they are all like like, you know, like
• 39:12
  you just mentioned the python, the transition from python 2 to
• 39:16
  3. And they all like, you can you
• 39:18
  can see them like Schreiber, it's like, oh no, they are not
• 39:21
  then you're not gonna do anything that will break
• 39:24
  backwards compatibility. That way not easily.
• 39:30
  So I don't think that's something that will happen.
• 39:32
  I also don't think the python will get like super specialize
• 39:36
  in one particular area or like, you know that.
• 39:39
  Oh. So now the python is only used
• 39:40
  for data science but is no longer used for, I don't know
• 39:43
  for terminal applications or for web development of it.
• 39:47
  I don't think that's going to happen either.
• 39:50
  I think it's just going to keep growing.
• 39:53
  I mean like I don't feel this and there's an area.
• 39:56
  All suddenly diminish and just disappear or something like that
• 39:59
  in Python. So yeah, like it's I don't know.
• 40:05
  I don't know if it's a good answer or not, but it's
• 40:07
  something that is not that I feel is not going to change like
• 40:10
  the diversity of things that you can do with python.
• 40:15
  Yeah, I think that's perfect. That's a great answer.
• 40:18
  One thing that I'm wondering about, I want to change gears
• 40:22
  real fast and get a bit more into the The Human Side of you
• 40:28
  and know about like because I've seen your you went all over the
• 40:33
  place, you were in Colombia or you started your life journey in
• 40:37
  Columbia and then you went to Saudi Arabia, you're now in, you
• 40:43
  know in Germany. There's got to be a lot of cool
• 40:47
  stuff that's happened along the way.
• 40:49
  I would love to hear about Saudi Arabia but I also just want to
• 40:53
  know like the main thing I want to know is how has your Life
• 40:56
  changed since. Fast API, like how have you as a
• 41:02
  person started to what has changed?
• 41:05
  I guess is the main thing and how Huh.
• 41:09
  That's that's a really interesting question like so
• 41:12
  much has to do, you know so much had changed even before first
• 41:16
  API, you know, the guy from Colombia and I'm from Colombia
• 41:20
  Latin America. That's why the accident for
• 41:22
  those that haven't noticed unlike hey, I ended up in Saudi
• 41:26
  Arabia and Bahrain, which is like, the, the Iceland on the
• 41:29
  side and then into y, for one year and a half, or something
• 41:32
  like that. And then now in Germany and now
• 41:34
  in from Germany, I'm working mostly for companies in the The
• 41:38
  states but remote is always like this weird thing like all over
• 41:42
  the place and like that that implies a lot of changes but
• 41:46
  after first API like I don't know.
• 41:48
  It's so so many things have changed its so it's very
• 41:52
  strange. It's really strange.
• 41:54
  It's of course is super exciting and very cool and like I get a
• 41:58
  lot of things out of it but at the same time is quite strange.
• 42:03
  Like you know, and you know like one thing that I think is
• 42:06
  particularly fun is I built this whole thing first API, blah,
• 42:12
  blah, blah. And I built this library for
• 42:14
  building commander in applications typer, which is
• 42:16
  like with the same style of fast API and then I built this
• 42:21
  library for communicating with with databases built on top of
• 42:24
  SQL command pedantic which is like kind of or M or something
• 42:28
  like that. But the thing that really got me
• 42:34
  more followers on Twitter was one tweet.
• 42:38
  That was fun. Um, and that went viral.
• 42:40
  The was just, like, mildly for her own and like that was the
• 42:44
  thing that was that gave me the most exposure, I guess.
• 42:48
  I don't know. It was very strange because
• 42:50
  like, I worked so much invested PA and I made a joke about for
• 42:55
  CPA and that's what was super super viral and like will
• 43:00
  brought a lot of people to see what I have to say.
• 43:02
  And like, hey, I want say a lot of fun things like I'm just
• 43:07
  talking about first API and pipe.
• 43:09
  I'm mostly worried with. Yeah and I think that what you
• 43:12
  said that famous tweet has made its rounds all across the
• 43:17
  machine learning and even software engineering internet.
• 43:20
  Because if you don't know and you're listening and you're
• 43:22
  wondering what the Tweet is. It was something along the lines
• 43:25
  of like a recruiter asked me if I had experience in fast API,
• 43:29
  and for, like, seven years experience, right?
• 43:32
  And I said, no, I only Built it. Five years ago or something like
• 43:35
  that, was it? Yeah.
• 43:37
  Is this a joke? Was that the required four plus
• 43:40
  years of experience in first EPI?
• 43:42
  And I couldn't apply because I built it that thing.
• 43:44
  One and a half years ago, that was the tweet and like the idea
• 43:48
  that under the end of it, it was maybe we should re-evaluate that
• 43:52
  idea that the skill level is the same as the experience in years.
• 43:57
  Which is something that I really, I don't know, something
• 44:00
  I feel very deeply and I struggle a lot with I don't like
• 44:03
  it when companies require years of experience because I feel
• 44:07
  like that's a a very, very bad proxy for skill level.
• 44:12
  Yeah. And for skills.
• 44:15
  Yeah, because you could have spent sitting around doing
• 44:18
  nothing for ten years at some big company or whatever and not
• 44:21
  gotten fired. Just just doing the minimum and
• 44:24
  then you have those ten years and you're getting the and
• 44:27
  someone who actually is hungry and wants.
• 44:29
  It is getting passed over because there's somebody else
• 44:33
  with ten years and this person who's hungry only has one year.
• 44:36
  Totally totally. And at the same time, at the
• 44:40
  same time, you have like I don't know this person that is just
• 44:42
  the was doing machine learning data, science or blah and then
• 44:47
  had to start doing front end at least start up.
• 44:50
  And then like, it was just a start up and it was just this
• 44:52
  massive amount of work that they became an expert in front end
• 44:55
  leading a tables to L developers at the same time.
• 44:58
  And like, you know, the amount of experience that this person
• 45:01
  can have in one or two years is just massive compared to what
• 45:06
  song? Someone that can have 58 years
• 45:10
  building the same exact back in the application with the same
• 45:13
  database with the same thing. Like, you know, not necessarily
• 45:16
  solving a new problem. Like the difference is the
• 45:20
  imbalance is very high in these things.
• 45:22
  So so yeah. I don't know.
• 45:23
  I feel like years of experience is above estimator.
• 45:27
  You know? It's of course, it's an
• 45:29
  estimator is an indicator, like, it has some kind of correlation
• 45:34
  but I think it's not great in particular.
• 45:36
  Now, the technology is advancing so much, maybe some years ago,
• 45:39
  when everyone was just writing Java with the same database and
• 45:44
  with the same things. And like, yeah.
• 45:45
  Well if it has, if that person has most more years, like they
• 45:48
  have to have seen more problems and like that's it, but
• 45:52
  nowadays, that that you have like new Frameworks, new tools,
• 45:55
  new features in the languages and new ways to do things that
• 45:58
  are faster and more productive than the ways that you had
• 46:02
  before. Like it's everything becomes
• 46:06
  more. Or more relative completely.
• 46:08
  So yeah you touched on an interesting point there and I
• 46:12
  think it ties in with the kind of Journey you went through with
• 46:16
  eventually starting to build fast API but People like,
• 46:22
  especially in the envelope space and the Machine only space.
• 46:24
  There are so many tools and libraries and Frameworks out
• 46:27
  there. It's hard to you can't become an
• 46:31
  expert in the mall. It's in fact, it's hard to even
• 46:33
  use them. All right?
• 46:35
  How like, how do you pick the winners for you or do you just
• 46:39
  stick to one and get good? Or do you think what's your
• 46:44
  approach for like, sifting through all that stuff, right?
• 46:48
  It's hard. That's a good question and like
• 46:51
  I Don't have it figured out and it's overwhelming and like, you
• 46:54
  know, like I see, I see in particular people that are new
• 46:57
  to the field like people that are learning and that they feel
• 47:01
  so overwhelmed and they feel like this massive imposter
• 47:04
  symptom, but they don't know that we all feel the same
• 47:07
  imposter syndrome and that we are all feeling like, oh, this
• 47:11
  is so impossible to learn everything, you know, I
• 47:14
  remember, I remember the ouster of pie torch tweeting about it
• 47:18
  like some monster or maybe one or two years ago or something.
• 47:21
  They saying, I don't know how to keep have to keep a typical the
• 47:26
  pace of the things that are changing.
• 47:28
  Like, how do you, how do you keep reading papers, reading
• 47:31
  these things like learning new libraries.
• 47:33
  It's I feel it's very difficult and very complex.
• 47:36
  My personal approach is like, well before I used to read a lot
• 47:41
  of blog posts, trying to compare one thing and the other, but
• 47:45
  then in the end in practice, I ended up realizing that this
• 47:50
  blog posts. In the end are like in one way
• 47:53
  or another bias to the point of view of that person and I have
• 47:58
  been finding more and more the things that I really want to
• 48:00
  optimize, for example, auto-completion and inline
• 48:03
  errors and like editor support. And these things are very
• 48:06
  important to me. And I want to be able to have
• 48:08
  that. And whatever comparison is done,
• 48:12
  by by someone else, is probably not going to have in mind, the
• 48:16
  things that I think are more important to me.
• 48:20
  So the What I have been doing like lately when exploring
• 48:24
  Technologies or things like that, is to try to follow the
• 48:28
  getting started tutorial of those Technologies and just like
• 48:31
  try to follow the, the very first steps and try to get the
• 48:34
  feeling of how it looks like how it works.
• 48:37
  And that gives me a lot more of a sense of like do I like this
• 48:42
  or not that much. And at the same time, if
• 48:46
  following, the getting started guide takes one week then it's
• 48:50
  also a Maybe not a great indicator because like maybe
• 48:55
  this thing is too complex and like, I don't know, maybe, maybe
• 48:58
  I can start with something simpler and figure and see if
• 49:00
  that that's good enough for the things that I want to do.
• 49:03
  That's also what I what I recommend with first API like
• 49:06
  when you're sinking like should I use first API or not instead
• 49:09
  of just reading and reading and reading what others have said
• 49:12
  that that you should do or not, I recommend just like reading
• 49:17
  that the first page like the, the main page of the website,
• 49:20
  just the main page. Age and following that tutorial
• 49:22
  and like executing it and like running it in the editor and it
• 49:26
  will take like 10 times or some 10 minutes or something like
• 49:29
  that. It's just just that no, no more.
• 49:31
  But that's enough to get a feeling of.
• 49:33
  Hey, I like this or not. Like do I like this approach?
• 49:36
  Or is it too weird to me and like I'm so used to working with
• 49:39
  Django to. I really don't want to try.
• 49:41
  This thing is too weird or maybe you can see like okay this this
• 49:45
  looks nice, maybe I can framework.
• 49:47
  That's I think that's kind of my Approach now.
• 49:50
  Try to try. To use the actual tool a little
• 49:53
  bit in some cases from the same examples.
• 49:57
  Now, I can before running the code, I can know that.
• 50:01
  Hey this kind of provide Auto completion.
• 50:02
  This won't be able to checks for the types because there's no way
• 50:08
  to annotate this type of thing just because like I helping
• 50:12
  getting so deep into this type of notations, on these things.
• 50:15
  But I feel like that the best way just to try to run the code
• 50:18
  try to test it locally a little bit.
• 50:21
  It and that gives a lot more sense of like, does it work for
• 50:24
  me or not? Incredible.
• 50:28
  So so good, man. I really appreciate this view
• 50:32
  point of me bringing. It's like the way that I see
• 50:36
  your framework, or what you are trying to do is, you're really,
• 50:41
  really deep on certain things that you really enjoy, or maybe
• 50:45
  it's not that. You're only deep in this area,
• 50:47
  but you really enjoy and you're, you're, you're shaped like, what
• 50:52
  you're passionate about. How can you bring that to other
• 50:55
  areas? And so, I think about when
• 50:58
  you're talking Using any scale and Ray and how can you bring
• 51:03
  auto-completion to that, right? So that is just that super
• 51:08
  awesome to see. I think my video might have gone
• 51:12
  out for a sec but but this is this is great.
• 51:18
  I think we can end it there. Unless Adam, do you have any
• 51:21
  final words? Just really fast.
• 51:26
  And I could honestly listen to you took away all days.
• 51:28
  It's obvious that you've done the hard miles is developing not
• 51:33
  the Deep technical expertise on the thing.
• 51:36
  That like the Mitre saying you really care about and actually
• 51:40
  when you get over the hill of difficulty there, you can start
• 51:43
  to apply that add value all over the shop, right?
• 51:46
  So that's really, really cool. Now look, thanks very much.
• 51:49
  It's been a yes, been great speaking Jeff.
• 51:52
  So amazing. Everyone out there.
• 51:54
  If you want to or I highly encourage you to follow
• 51:58
  Sebastian on LinkedIn you'll just see him posting all
• 52:01
  pictures of him. Traveling around the world and
• 52:04
  it will make you very jealous for meeting with all kinds of
• 52:07
  cool people or follow them on Twitter will leave the links to
• 52:10
  those below. And then you can check out fast,
• 52:13
  API, will leave a link to that he's also on.
• 52:14
  Get, he's got an incredible GitHub that you can check out.
• 52:17
  We got all the links for you in the description and that's it.
• 52:20
  I think we can enter their thanks Sebastian.
• 52:22
  Thank you very much. Thank you.
• 52:23
  For having you know, like I had so many doubts that I will be
• 52:27
  able to break anything useful to envelopes Community, but it's
• 52:32
  been fantastic. It was very, very fun chatting
• 52:34
  with you too.