Transcript: How France Ditched Microsoft - Samuel Paccoud

Whether it's scaling an application, deriving insights from data, or getting results from AI, Six Feet Up helps you move forward faster.

See what's possible at sixfeetup.com.

Hi, welcome to another episode of Django Chat.

I'm Will Vincent with Carlton Gibson.

Hey Carlton.

Hello, Will.

Welcome.

Thank you for having me.

Happy to be here.

So even in America, the the work that you're doing has raised quite a stir of having open source uh software to replace Microsoft and Zoom and other things.

Um perhaps you could tell us just broadly uh what you do day to day and then we'll go from there.

We are attached to the French Prime Minister, and uh my day-to-day work is to build a set of applications which we call La Suite for daily work.

Uh so they are there's instant messaging, uh emailing, uh video conferencing, organizing sharing files and drives, uh, collaborative editing, spreadsheets.

We even uh added an AI chatbot recently.

So everything for the daily work.

Um and we yeah, we do it for uh state agents, for public agents, but uh we do it in open source, so it's there's also a community growing around what we do.

Yeah.

Also for the for the government?

And before that, I was working in the higher education uh sector, also doing Django with uh a lot of um open edicts, if you know open edicts.

Yeah, yeah.

It's one of the yeah, that's one of the big Django projects.

Um so yeah, I worked for six or seven years in in uh France Université Numérique.

So that's the equivalent of Dinum for universities.

We we we provide uh digital services for universities, cloud services.

And then I think um the history is that during COVID, is that when there was effort started to make these services available across government?

But I was working uh in universities, I was working before COVID.

When the COVID started, we were the first team to spin up a Jitsi instance, and we had also uh collaborative editing.

We we very fast we tried to give tools to our agents, which were suddenly at home and had nothing to work.

So um, because we were able to very quickly put up services online, um, people started to notice us and send us more money and say, okay, we need more of this because people are remote and they need to work.

So I think that's when people realized that uh open source was where people had a very fast response to these issues, and and they started giving more money to open source projects.

Uh while I was in France Université Numérique in DNU, they were also start, uh they were also able to propose tools uh and with open source, but these were standalone tools.

We didn't have a full digital workspace interoperable with interoperable applications, and so that was the the next step was to say, okay, now people are after COVID, people started to say, okay, we need to switch to we need to switch to Office 365.

And like public agents, they didn't have these online tools.

They were just working on their laptops.

And after the COVID, people started to want to move to Office 365.

That was that was raising a lot of questions in the in the French public sector to say, okay, are we really sovereign if everybody's moves online?

We started to realize that these daily tools are not just uh they are very important because all your data and all the work you do uh is done in these tools.

So they are really, in fact, they are not just just uh detail, they they are they are really core to to your sovereignty, to your data and stuff.

Yeah.

There's cost and the digital uh sovereignty because in the public sector we we we wouldn't have enough money to pay all the licenses for all our public agents.

So we we were they were starting to say, okay, let's give Office 365 to only the maybe the top level managers or yeah, they were starting to have this kind of uh decisions, and no, that's not what we want.

Uh and because there was also a big problem with uh sovereignty, then they started to see a solution with open source.

Uh but really, frankly, when I arrived in Zim, nobody thought that we can do it, and I myself didn't know what I was going to be able to do.

And that's when we started to have top one in uh Hacker News uh with docs and then visio later, and then we start to believe in ourselves, saying, like, okay, we are able to do it, and we can maybe it's easier than we thought.

So maybe maybe it can be done.

What sort of backing do you have?

So when I when I took the team, the budget was it was about like between five and six million, maybe.

Everything was used to pay licenses.

So I had to work nights, weekends to start to to start docs and video.

We started it as a student project, we did it on our spare time.

Um and then slowly when we did top one in Iconews, people say, okay, maybe we should give them more money.

And then so today uh last suite is is roughly is around 10 million budget, maybe.

Okay.

It's about 80 people, uh, 35 developers, uh mostly freelancers.

Yeah, yeah, that's that's the figures.

Let's have you know, be able to put a team of that size together.

That's not a small team, that's a big group.

Like when you you know, the state state alone in France is 2.9 million agents.

Yeah.

And if you consider the whole public service in France, it's uh six million agents.

So the potential for these open source tools is is huge only in the public sector.

And that's why it's uh strategic to it could it can be done in the public sector.

It it can't be done in the private sector by just a company starting to starting and say, I'm going to uh make a suite with open source and spread it.

This this wouldn't work.

So we thought that okay, uh the public sector can make this happen.

Okay, cool, cool, cool.

Because team licenses, especially for enterprises, are I mean it must be millions of dollars that's saved separate from the costs invested in your team.

So yes.

That's the word we put it.

But it's more complex.

Because did we introduce the suite?

Because we we we we know it what it well now, I know what it is, that's why we've invited you on.

But did we really introduce for the listeners what the last suite numeric is doing now?

I don't know if you want to go into each application.

There's there's also a a a core of last suite which is growing with which is based on Django and other applications which are not.

So we can maybe focus on on the Django uh applications.

But but they they are there were some existing tools when I joined.

Uh the first one is uh tool called CHAP, which is uh basically a government instance of metrics.

Uh so that's uh 400,000 active users, uh monthly active users as we speak on on the metrics uh federation for the French states.

Yeah, that's

Okay.

We have a private federation.

It's quite successful also in uh uh how do you call it regalion uh environments in governments because it's encrypted that it's encrypted and distributed uh nature of metrics.

Okay.

I assume it also has to be like servers based in France.

You're not using AWS in you know Washington Washington, DC?

So so it excludes uh American uh led uh clouds.

So yes.

Uh but we have several instances.

We use public cloud with this certification of second cloud, so it could be the French uh hyperscalers are uh like OVH or Scaleway.

If they have this certification, we we can use them.

But but we also have internal clouds, internal to the states where we can have specific instances with more security.

And that's the interesting thing with open source is that once we have the product, we can deploy it tens of tens of times uh in air gap, in uh classified environments, in open environments, and it also spreads uh uh outside of the state.

So that's interesting.

Uh yeah, it compounds.

Um is like I think once you've built the foundation, there is a real chance to if you if it makes it easy enough to install and uh people can run it themselves, there's a real growth opportunity there.

The market and the cathedral and the bazaar type you know approach could really apply here.

The the especially on docs and visio, the uh like Visio for was used by gamers because in Visio you can share several screens at the same time, and you can stream it to RTMP, so you can stream it to YouTube and that's kind of so gamers started using Visio.

It was packaged for NixOS as well, so very easy to deploy, and and it's funny because the people who packaged it for NixOS they made a fork activating all uh some features for gamers.

That that was the first community that started using uh Visio.

Um and then it as we speak, there are I don't know how many instances there are, but they're they're quite they are it's growing.

And yeah, we have people who developed a full mobile application last month, and that's it's coming.

Like contributions are starting to come.

Yeah.

When you want to build a digital workspace, you have to build a lot of applications, and yeah, we need to be big, and we are not that big either.

So we hope we we need the community to take on this and to help.

But we also we are like myself, I I've been working on Django for 10 years, and we we have about 10 developers who are really experienced on Django.

And so we reuse what we do a lot.

There's a joke in LaSuite saying that everything is a fork of docs, and if you in the case of Drive, Drive is literally a fork of docs with just a small change in the front end, we replace the editor by a file browser, you know, and that's it.

That's that's a tree, documents and with access rights.

It's really, really close.

So we we manage to because Django and React uh are uh we we master our frameworks, we we're able to reuse a lot of what we do.

And yeah, we try to to cover the whole digital workspace with the with the resources we have, which which are growing but slowly, and trying to embark the community and yeah, be smart too.

Like have you must have been tempted by Django Ninja and maybe some other tools.

Like how is how is Django Rest framework holding up for you?

I guess is the question.

And we had our habits and the way we do things, and we we focused on what we can build as fast as possible.

And we didn't take the time to like Django Ninja when we started, I was I was looking at it and I hesitated, but it was really the beginning of big uh yeah, I couldn't find like a big project in production that would make me feel really safe to use it.

So I I just went on with REST Frameworks, which we were which we were using for several years.

Then lately I I saw, of course, I saw Django Bolt, and I'm really really excited with Django Bolt.

Uh but you know, there would be a lot of rewrites.

Um I think we would love to stop and make this move, but it's also not like it's you don't need to.

So if you just focus on what you ship and what you your impact, the tools and your users, tools are it's good enough.

It's we master it, we know we can do whatever we want, and we it's not rocket science.

What we do is really it serializes views and models.

We've been doing it for years, know how to optimize it.

We don't even have cash yet on our docs or drive.

And yeah, we know we have really room for for optimizations before we need to go to Django Bolt, for example.

But the truth is, uh then they say, Oh, Python is so slow when you choose Django.

But I mean, if you when you build these applications, you know that your your the speed of your application is really most of the time it's just in the database or the way you make your query.

1000%.

1000%.

It's just like people don't know what they're talking about.

Oh my god.

In France Université numérique, we had a video, like a YouTube platform uh equivalence, and we we had three million users on it, and we had only one pod because like it's just serving video tokens, and uh so everything was in cache, and so we could serve like so many requests with just a pod.

So yeah, sometimes it's just uh not where the paint is.

Perfect.

We love we love the URM in Django, so okay.

One of the challenges before was getting adoption was login, and maybe you could speak to uh introducing that and why that I think led to a lot more users for you.

They are in each ministry, they they are in small entities, organizations are everywhere, and it's nothing like hierarchical.

There's no one who can say, let's do that, and everybody follows.

It's it's a state, so it's designed to be complex and and have independence between teams so that not one person can decide for everybody.

You know, that could be creepy.

So that's also you need to convince people that your tools are good, uh, and to convince them they need to have access to your tools.

Yeah, right.

And that's that could be really, really uh a headache.

Uh so but but we we are we were able to put up a SSO which has every everybody behind one button and they could log in in one click.

So that that was really, really important in in our distribution strategy for our tools.

So to give it a try, kick the wheels.

Yeah.

Oh yes.

Uh we get inspiration.

We in France we have France Connect.

We had France Connect for uh for end users uh with like 100 million connections per per month, and it was working really well.

But so we we forked it and used it for ages.

Okay.

I'm wondering um whether other governments in the EU are picking it up and joining in.

Uh but but what we don't have is like uh we don't have a common response.

Like everybody say, okay, let's do this, and everybody, you do this, I do that, and that's it.

We're not able to do this.

So I would say it's important that everybody that the concern is raising, and open source is quite uh uh yeah, it's it's seen as a solution in several countries, uh, especially Germany, France, um, the Netherlands also is joining us on this.

Um so yes, it's it's there's a concern, but I would say we are not there where we where our efforts compound not yet.

Try to work together.

But for example, with Germany and and Holland, we and the Netherlands, we we try to work together, but uh the approaches are are quite different.

So the yeah, the I wouldn't say that that uh yeah, that we're very, very successful to to add it add our efforts to compound.

Yeah.

I mean the the European project is still quite nice

Like if if if you m if one country pushes something forward, another country pushes something else forward, eventually it everybody gets aligned behind what works.

So it's also good that everybody's trying a different approach.

And if you hope that what works gets uh yeah, gets spread.

It's that it might take slightly longer to get off the ground, but hang on, the the Dutch perspective there brings something that we hadn't considered to the table.

Yes.

Like you know, like you're you're in the public sector, so you have a lot of pressure, a lot of lobbies.

When you succeed in something, the people come for you, they can, you know, you can you can lose some battles, you have to continue fighting.

Okay, so you have to hope that that you have a positive impact on on these things, but it's it's never never ending.

Okay.

Um it's I would say it's getting it's gotten quite easy on docs, on on the projects that had the most uh the biggest community.

Yeah.

Um so we we make it um containerized, we have Docker Compose.

If you go to Docs, you pull the repository, you make make Bootstrap, normally it works on your laptop.

Like if if you're a developer and you have Docker on your computer, it should be working within minutes.

Okay.

Um on develop, um, on going to production, we focus on Kubernetes.

Like we provide everything for Kubernetes deployment to production in minutes, like ham charts, uh, we propos we have official images and everything.

So, and documentation.

Um, so if you run Kubernetes is easy, if you don't, okay, we we you're a bit yeah,

But if you master Django, it's a bit too hard.

Like it's yeah.

Okay, good.

Um and and other projects we on other projects who are more recent or on which we did less community work, maybe the it's uh there's less com less documentation could be a bit harder.

So still work in progress to really make it easy to deploy.

Right, okay.

I mean, but yeah.

I'm trying to deploy this, I'm trying to do that.

And yeah,

of course.

Um can open issues.

We we're quite active, uh, we're quite responsive on all this.

Um is that obviously we build uh applications for scale, like like we're talking million users.

Yes.

So we make some choices, which are like we make object storage first, we we make choices for scale, uh OIDC first, everything.

So if if you have a very small company and you just want to have a VM and pay the VM one euro per month and you want to install everything on it, you need to be skilled because you will need to have a OIDC provider, you will need to have uh object storage.

A lot a lot of people they they will if you want a very small instance, they will realize that uh it's not it's not what we focus on.

So if you don't have S3, then

What's stopping you?

Maybe it's an application that won't scale or an AI initiative that just isn't delivering?

That's where Six Feet Up comes in.

With deep expertise in Python, Django, and AI, they solve hard software problems, modernize platforms, and get teams to market faster.

Their impact speaks for itself.

Automating healthcare pipelines for hospitals, helping NASA explore Pluto, building severe weather prediction tools, and applying AI to connect farmers with vital crop data.

When the stakes are high and the problems are hard, six feet up is the partner that delivers.

See what's possible at six feet up is the partner that delivers

But there's an interesting contribution opportunity for people, like the people who want to scale it down to come along and you know.

I don't know if you know this project.

You know host?

No.

Yeah.

That's like uh yeah, maybe it's French.

It's it's quite it's quite used in free software communities.

It's like uh a box on which you can self-host a lot of applications without doing like heavy sysadmin.

You can click and install your so they have put Visual, they've put Gris docs, you can install it with a click.

I'm not I'm not in favor of these approaches so much because sometimes you make it look like it's easy, but when you run uh an object storage cluster and it's not so easy, you know, you still need skills in case something goes wrong.

So yeah, if you hide too much complexity, if you hide the complexity too much, sometimes people can have problems.

Can you dig down and work out what the separate

parts are?

So I would advise that you open a bucket on a provider, and if you're if you want sovereignty of your data, you can encrypt, although you put it on a bucket on on the cloud, you can encrypt your data on this uh S3.

But at least you don't have to operate it.

You're not scared to lose your data.

So that's my advice.

It's cheap to have a bucket.

So yeah, okay, cool, cool, cool.

Um, I actually had a conversation with the developer advocate there, which is because I was surprised, I I hadn't heard about it until Heroku shut down again.

But they have proc files, they've been going 10 years.

I believe they work with the government a little bit.

But it's a very, very easy-to-use option.

So we can deploy to Scalingo.

Uh yeah, we have some instances of Scalingo.

There's also Clever Cloud in France, which does a bit the same thing.

It's a path as well.

And they have uh hard tools, detailed hard-tools of how to deploy docs.

Yeah, that's a good option as well.

But maybe more expensive if you're really looking for cheap uh hosting.

So yeah, you basically need solutions for all the profiles.

Yeah, yeah.

You need all of those eventually.

If people want to come in and add to the docs on a deployment story or something like that, they could

do that.

So that's public agents, uh the the repositories are we we are there's no open governance at the moment.

But that's one of one of our main objectives.

We are we when we say we want to do digital commons, we mean it's we mean that we want the governance to be shared.

We want to slowly, we want to just be one of the contributors.

Uh so we we were hoping to bring, we are we are still hoping to bring uh other countries, uh companies, integrators, hosters, uh to contribute and slowly take over governance with the community.

We are not there yet, and we we frankly we don't know exactly how it will be done, but that's where we want to go.

I think docs is to be the head and will be the first repository to where that we will really open, most probably send it to a foundation and really look for an open governance.

Um there are some communities, I don't know how much I can say, but there are some communities in the the open source, in the free software uh communities that that are putting offers to DNI to take the code and start to organize this for us.

I don't know what it will what what will happen.

I frankly don't know.

But I really, really hope that this goes through.

You know, it's a cool thing you've built and it's you know open source, but then to give it an open governance model gives it sustainability into the you know medium and long term, hopefully.

So what are the best practices, what are examples, what what has been done, and yeah, we we are at the moment we are looking what's possible and yeah.

That's a cool so how how did you end up in this you know great situation?

Like you're be you're working on what's a really cool project in a kind of socially worthy uh thing at the cutting edge of technology doing some really good stuff.

And how did you it's a kind of dream job now to roll into it?

No, no, no.

I don't know if you're on LinkedIn, but it's quite it's it's quite uh I don't know how long you can hold on this kind of job, you know, it's quite active and quite exposed, like yeah.

So but when you succeed, when you succeed it's good, but um not a dream job.

Um it's it's uh what I was looking for, so but that's not easy.

So maybe if it was a bit easier would be the dream job.

Uh well, how did that be?

Yeah, I I ended I ended up here by because I uh like I was like I said, I was a Pango developer and the f the first I became a developer to to grow a family.

Like I was looking for a job that would really accommodate taking care of the family like uh during the day.

And I was when my kids were small, I was coding at night.

I was I wanted something really flexible, so I went, I I came to coding quite late.

I was like uh past 30 years old.

And before that I was working in uh international commerce.

So it gives me a profile, a bit double profile of uh learning coding late and and having a business, international business uh experience before.

And so that that's a strength that that really uh allows me to navigate in the complexity of public sector and yeah, open source stuff.

And so you come with a mission, like you you you see something that's in the goes wrong in your country, or something like digital digital sovereignty was really an issue.

I've I've always been sensitive to this, like like really um the uh predatory, no, like predatory software.

Uh yeah, but that was always a big issue for me, like really from young.

And I think I I came back to this with the excuse of building a family and having a more flexible job, but I it was in my head that okay, that's that's one thing that's important for me.

So yeah.

And when when coming to it, um yeah, you the the I started to learn coding for 10 years as a freelance and slowly moved to really to open source, and it was it was progressive.

When I when I arrived in France Université Numérique, I was I was coming for open education more than for open source.

I I didn't really I was not really in the open source communities, I was using Django, but not really I was not really a free software advocate in France or known for this.

Uh and then from working in open education and slowly, slowly I I started to really know where uh to start having strategy of how to do it, how can we make it happen?

Um and the strategy I'm applying at the name is quite controversial.

It's still quite controversial, even in the free software uh communities, uh in the private software.

So because we succeed, people start to see, oh, maybe that's something interesting the way they do it, but it's controversial.

Like we are in the middle of the open source movement, the free software movement, the and and the political things.

We need to convince everybody that this is how you get impact, this is how you keep control of your software, of your data.

Uh yeah.

So our success gives us a gives this strategy a voice.

Yes.

I can elaborate a bit more if you want on this.

No, but

You have to, you know, get documents out the door.

Well, when when you succeed something, then people start to think that okay, they they misunderstood something.

If not, they will think they are right and you must be wrong.

So when you succeed something, it then we can talk again that okay, see this thing you said that we I was going to fail, but in fact I succeeded.

So that's why I succeeded.

Do you agree with this?

So now I I propose that we go this way.

Yeah, okay.

Uh so that gives us an opportunity to maybe build what like you know, it's very interesting.

The uh some people come to see me and say, okay, we were trying to do this for 20 years in France, we didn't know how to do it.

Please explain how you did it because we want to apply it in other sectors.

Like the like what's the recipe for this?

And the recipe is you know, there's a big debate here on the make or buy.

Should you when when your government should you be coding?

Like people will say, no, as government you shouldn't be coding.

You should just be uh defining the specifications, the interfaces, the standards, and that's it.

Your job is done.

So maybe in some countries that's that's what the the role of the state is, but in France, the the state sector is much, much bigger.

Like yeah, we were saying, like six million agents.

Um and I don't think we and and the uh history proves that when you you're not coding and you're trying to regulate stuff and declare standards and specifications.

If you're not coding, you have no idea what you're talking about.

Like, you know, AI is coming.

And I also have no idea before I start coding what it should be.

Now we we are working on interfaces.

I was mentioning that now that we succeeded, some applications, we want to make them interchangeable so that when people sell digital workplaces, we we have a standard that we want to call Open Bureau.

And if you comply to these standards, you can change a drive for another drive, a video conferencing for another video conferencing, and it will work in your digital workspace.

And when we started building this, we realized how difficult it is.

Because it's not just saying that's web dev or that's FTP or whatever.

You you can't just say that.

If you don't trust your drive, doesn't trust the video conferencing application, you need to go down to the scenario, you need a token for, and then you need to know what you will allow one application to do in the other application on behalf of the user or how long.

A little bit like the S3 policy, you know, S3 policy, you can craft uh S3 policy, but you will say you're allowed to push a file at this place, at this path, this maximum size, this type of file, for you have 30 minutes to do it, and then that's it.

Yeah.

And so you go down to the scenario and you say, okay, this is how it should work.

But if you're not the developer doing it, how can you specify it?

So I'm trying to tell people I can't, if I don't make, I can't buy.

If I don't make, I can't specify.

That's exactly it.

I'm so I'm I'm obviously from Britain originally, and the the the the opposite approach was always taken.

Let's um let's just outsource everything, let's do it all in the private sector for efficient and you know, and you know, that like all things, you can swing the pendulum one way or the other and it can go too far.

But the reality is you can't specify software with sufficient clarity to be able to outsource it totally.

And so you end up with you know the the the history of British IT projects over the last 30 years has been one of overspent and um you know missed missed specifications and then you know budget doubling and all the rest.

There's yeah, yeah, that's everything.

But it it's also it's a kind of logical impossibility to specify the software up front until you're in the in the weeds writing it, right?

It's something that we have at the smallest scale with writing our own applications, but it must apply a hundredfold at government scale.

And so maybe they're right that it's important to that the public sector is not going to do everything on its own.

Yeah, and and I always thought of like if I succeed, if I want to succeed last week, I cannot just succeed in the French state.

If I do this, in five years someone will come and say, how come French state has its own digital workspace where everybody's doing something else?

So the the project can only be we are building tools for our agents, but building it with open source, I mean, public money should go to public code, and our ambition should be to make an alternative to what the hegemons we have today that that uh that pose a sovereignty issue.

And so we want an open software, we want open software in Europe, we want diversity, and this cannot this is the only success we we can't have success if we don't manage this.

Um and so that when you say that, then your your project has become 10 times harder.

So yeah, uh and it's for that's why I'm doing a lot of politics because I can't just succeed by coding stuff that my users use.

I need to code stuff that the community takes, that other government takes, that private sector takes, that uh I don't know, everybody should be able to get value from it.

Yeah.

Yeah.

You're asked to to boil the ocean.

You succeeded in one thing and then nine more problems, and and uh yeah.

It's a success problem, but still a problem.

So we we we are in a better place now than two years ago.

We still have a lot of work.

Um but we we where would we would we be if we hadn't started it uh two years ago?

Yeah, okay.

Like when I joined, there were already two ministers in France who had uh asked for derogation to use Office 265, and all the other ministries were asking derogations as well.

And now no one is asking derogations, and they they give us part of the money to build a suite.

So we are in a better place, and we think if we continue, we are allowed to continue and grow out of the state, uh it's always something that will be good.

Um so that's fantastic.

Can I ask you, um, you know, was it just your historical background that led to Django or

you know it's you know why Django?

Yeah, I think he was doing Python and yeah, I think why I stick to Django is maybe like I was saying, my background was I was I had an engineering background, but not uh developer.

And I and and so I I wanted to make things done, like I wanted to make things, and I was less attached to the tech than to what I could do with it.

And then that's what I like in Django, like you know, the the efficiency, batteries included, you can ship some things.

I I really love the also I I always liked the Zen of Python, clean code.

Um I feel that's that's more than the product, like it's more than the code, is it's the community.

Yeah.

I I don't know.

Like in France, PHP has remained uh bigger than elsewhere for many reasons.

So Python and Django community is a very small world where everybody knows each other.

It's like really a family spirit, and so I was attached to it like very fast, you you know, more or less.

I was a freelancer, so more or less you know everybody, and that there was really like a quality approach to code, and that's how we existed.

Like the Django community 10 years ago existed by the being making things differently than the rest of the world.

That's what I liked.

And now it's the boring tech, it's really it works, it will work in 20 years.

Yeah, the solution for every every problem is it's uh already solved in an elegant way.

I mean, that's not the that's the second best tool for anything.

And AI AI for me is making the choice of Django even more obvious because it's the same engineers, and you see a lot of people from the Django community who have now moved to the AI community.

When when you go to Python, I went to EuroPython in 2021 and before Chat GPT, and yeah, it was already everybody was talking AI.

You have you have a huge community around AI in Python, so it makes a lot of sense to do Python today.

And together with Rust, I think it's really uh also it makes it maybe future proof.

It could be discussed, but I'm still I still feel very comfortable with this framework.

Okay.

It it's very strong.

I would say it's number three at just in in my head in terms of public-facing companies that are doing cutting-edge AI stuff.

I think so, because all the people you mentioned, they are from the same school in France.

So it's there's a very yeah.

You you had mentioned at the beginning adding AI enabled features to La Suite.

Um, how are you doing that?

Is that loc local models is that like what models are you allowed to use I assume maybe not you know anthropic that's a sovereignty question no

could we have uh also live live transcoding uh LLM LVM any type of model we could run but our approach to AI is we know our limits we we're not AI engineers and we we try to use AI for our use case we try to not invent a use case and we take an existing use case something that really works or already works obviously and we tr

y to bring it to our tools so for example in Visio we we will bring transcription live transcription uh asynchronous transcription and live transcription summary of the transcription um in docs we have basic uh editing uh uh features like uh text generation features um and in files in the drive we do a rag we we try to index to make um semanti

c search we we try to work on OCR on uh you know indexing any type of document efficiently uh so these these are all use cases that are mastered and that really you just need to run a model and make a little bit of identical a little bit of uh front end and make it work um that's the approach we have now um as on it works but we could be better ye

ah we we would need some help we could we could take some help on this part of the out of our products but

but we we only use open source models okay

security um and they're good enough a lot of the time I think it's getting

of the tools so if your people are are if everybody is in Office 365 then you don't choose the model that they'll be using.

So why do you do all these AI projects?

Because your people are everyday the tool which you don't control you can either pay the AI feature or don't pay it and you don't get it.

So um we we believe also that because we do the infrastructure stuff because because we do the structural stuff and control our data and our users, we'll be able to bring the use case and to bring usage to the models that we run.

And something that we see very few people manage uh doing and that we are able to do is that our applications we have OIDC to authenticate our users in each of our applications and not only our application any application in the French state which decides to connect to our SSO is in this federation in this OIDC federation and we use one spec of the OID one part of the

OIDC spec which talks about server-to-server uh uh authentication and so when someone is logged in on one app the server of this app can make a call to another app on behalf of the person who is logged in with the special access rights which I was mentioning to secure with we can get a token for a special use case.

And so we are able to tell any application in the French state would get the right to push a file to our drive on behalf of the users that is logged in.

And so this also applies to our chatbot and to our global search index.

We like I know Google did it 20 years ago but with French agents still don't have didn't have it.

So now we have a global search with semantic search with AI indexation and whatever whichever application which is on our SSO can push data to this global search and then users can query this global search from another application on behalf of the user our chatbot has tools that can interrogate this also this search with the the access right of the user and this

very people know how to do that so you I mean safely with with keeping control of your data and being allowing your users to query data live and do uh yeah I don't know if I explained it well but no

They used Algolia which is that French as well I think

very basic and on on the front end side for queries we use this resource server this OIDC resource server specification and then it's really easy because you just need a field for each document that's being pushed we we have one field with the users with the list of users who have direct access specific access to this document we have another field for any group that have access to this document and we have a few flags to say if it's public reach or it's not not

it could be not restricted to these people or to these users or groups but it could be public.

And with just this we're able to secure access and make a search that is restricted to what you can access.

So it's really actually really basic and the resource server part is a specification we have a few classes of a few hundred lines of Python to to check those tokens uh with the OIDC provider but it's really not rocket science and that that's really

Right it's very cool even a even someone who doesn't know how to code can that solves a big problem for them

And and if you don't know how to do this you know I I see my some colleagues who come up with oh we have this AI demo it's really cool and then tell them okay how are you going to plug it how are you going to give it to people and plug it to their tools and then that's becomes easy easy it just becomes much harder.

Okay.

People are asking for teams.

Basically for chat for Slack.

The Slack is real Slack.

So that's something people ask um and that's something we don't want to leave unanswered.

So yeah, that's that's something that's next for the suite.

We don't know yet what technology, and maybe that's one of the weaknesses in Django to uh everything webs okay, it's uh you know at scale.

So we don't know yet whether we use matrix, whether we'll use something like APUP serving another language.

Yeah, we we're not sure yet.

We know Django is not the best tool for that at the moment with channels.

Um and maybe I missed something and I'd be interested if someone thinks that there's a way to do it.

That's um, I mean, you know, there's um so the the the tool I'm looking at at the moment actually is uh called um Hornbeam, which is um written by um Bemoir Um escapes me this moment, but he's the been the maintainer of Gunicorn for Unicorn for a long time.

Oh yeah.

An Erlang server um with Python integration um as well.

I should

I saw him in Fosdem, I met him in Fosdem.

Um it's a server called Hornbeam that I've been tying up very much.

We are analyzing the first box proof of concept is uh end-to-end encryption on all our tools.

Okay.

So yeah.

And once you have a key globally for the suite, any application can ask for to encode something in the browser, and then when we did our application, we thought of we thought that we knew that we wanted to go this way.

So docs, fichier, uh drive, everything is is uh already planned to be uh and to an encrypted like like the server is dump.

Um so that's one thing.

We also have a big challenge to of course continue growing the community, continue getting more active users in the front states, because that's how people evaluate our work.

What is what's the entryway to Nasuite's community and to get involved?

Maybe you can give the link.

Um from our GitHub, you have you can see the different projects and on the readmes, you you find the metrics channel, you you can yeah, you can open issues, you can talk to us like this.

You can find us on maybe on on social networks, Mastodon, um Twitter or LinkedIn.

You need to make a social network for the French state.

Something that's not

some of those.

It's uh European infrastructure.

Oh, I saw this because you you gave a presentation, I'll put a link with a uh Leia German colleague, I think.

Yeah, but that that just started.

The edict was created and end of December 2025, and the the new director has just been appointed and will start working.

And so the edict is here to organize uh investment in commons at the European level, and there should be big project, big financing coming and that could be piped to this uh consortium.

So we we hope cross fingers that uh but the European Commission has announced that they want to push open source at the European level.

So the edic is uh is a good uh vehicle for this, and there is a strong, strong uh will to push uh social network, uh open source and more.

Yeah.

There there is this will.

I'm not sure it will be our team.

I don't think it will be done at the French state level.

Right.

Yeah.

similar space.

Do you know NGI?

What's that one?

Okay, I'm not sure.

Um that that's okay.

Sovereign tech fund is to to finance the maintenance of core open source libraries like FFmpeg or SSH, these kind of core projects.

And they they give they just give them money and they evaluate their success uh by the impact on everybody in the companies who have dependencies on on these libraries.

So that the project is really to make it more stable.

Um Nets is financing very small projects that uh most of the time it's like a group of two, three developers who say, hey, I'm going to code a new object storage, decentralized object storage with Rust.

And they say, Oh, okay, we give you 50k for that.

And it they literally take two days to decide to give you 50k, and then and then they they are really it's done by developers, it's done by really technical people.

So they they really look at your code, pull your pull uh they look at your pull request, they're going to really a technical evaluation of what you did.

Um and so it's a really efficient uh way to finance uh innovation, open source innovation.

And a lot of projects that we use in last suit are come from uh NLNET.

So, for example, in DOX, the editor, the block note editor, is financed by NLnet.

It's a Dutch guy who was financed by Nnet.

Um there are there are many uh small projects that come from this financial system.

I see that I think their budget is a lot bigger than the sovereign tech fund.

It's yeah, maybe ten times bigger.

So Carlton, you should apply.

Definitely.

Maybe, maybe, maybe.

Yeah.

For example, like let's say social network, let's talk about Mastodon or whatever other software that's here or platform that's here and that could grow or that could that would need financing.

Uh the edict could be doing this.

Like the edict needs to, it would be cross-border projects that need to get more traction and get bigger.

Digit digital workspace is uh is one cloud infrastructure, is another key key sector to boost.

Uh AI maybe, but AI is capital intensive, so maybe we will burn a lot of money and have less impact.

So I I would say digital workspace is one first thing to succeed, to have a strong, you know, strong uh structure, um, know how to take care of your data and host your projects.

Yeah.

Okay.

Um just as we finish up.

Um maybe I think do you want to go first?

Is there uh a book that you wanted to recommend, Samuel?

So I I recommend it to anybody who's talking to me uh about take more than two minutes.

Um that's the Mikado method.

Um I don't know if you if you read that book.

Carlton, you must have read it.

You ever read it?

It was a fork of OpenEntX, and it was really OpenEdicks is a huge code base, and I had like thousands of commits of fork on OpenEntX and something like really, really stuck in production with three million users.

And you I came to this project, everybody had left, like all the developers had been fired, and I was like alone with this huge code base.

And I thought I'm really good at Django, you know, but when I look at the code base, I don't understand anything.

It's not the best practice code, like OpenLinux is has come some really, really complex parts.

And um and I was really depressed about two.

There were two CTOs who resigned before me.

After taking the job, they stayed two months and they resigned.

I was the third one to try.

Yeah, and I told my brother about this, and he said, okay, you should buy this book.

This book is awesome because they he you know the game, the big adult game where you you you have all the sticks.

Uh you can't move one.

You can only

move one.

Yeah, okay.

Yeah.

And then you try to take one without moving the others.

And that's what you're trying to do with your code.

You you try to find the stick that you can take, and and then once you manage to take a stick without breaking all the rest, you build, you build it out of the monolith.

And so that's what I did.

I did for six years.

And what it where it's really interesting is that from day one, you're able to take a Mikado and prove to your management that you're moving ahead.

Like when they hired me, they said we are not able to change, come up with one feature in one year.

We are stuck with this code base.

If you touch something, it breaks.

Yeah.

So once I find a Mikado, like one month in the job, I was able to say, okay, you have a new tool, it now you have a new feature in this tool because I built it a site.

And then we kept on taking Mikados and eventually rebuilds, like, of course, it becomes a micro um uh microservice infrastructure because you take your Mikado and make services.

But from day one, you have impact as a CTO, you can prove that okay, now I'm fluid.

Although you don't understand a word of all this monolith, yeah.

So this book is game-changing for someone who really comes to a big legacy code base and to show what they can do.

I'm like, oh yeah, yeah, I need this book.

I'm gonna

So good.

And it's worse if you don't have tests.

If you have tests, it's maybe too easy.

Carlton you can always write a Selenium test around the outside or something, right?

You you just roll back, you just yeah, you just git reset and then you start another way.

You start another way, you start they have a whole technique to say, oh, maybe this is a good way.

Yeah.

Yeah, good, perfect.

Okay, I'm

Okay, brilliant.

I saw it on Kindle.

I saw, okay, I'm gonna get that.

It was the the the the Millennium series trilogy, uh the Millennium Trilogy.

I'm fairly sure that Paolo Melacorli has mentioned this as his sort of one of his favourite books, and I saw it and I was like, oh, it's on special.

Right, I'm gonna I'm gonna get that.

And it's it's good.

I'm enjoying it.

I'm just having a couple of you know, a couple of quiet, uh, a couple of quiet weeks away from the tech book, so I'm reading those.

Um yeah.

Yeah, cool.

Sorry, go ahead.

Can you put it

Sorry, I don't understand.

Yeah, I didn't get the title.

So it's

Yeah, but yeah.

There's a trilogy, and the trilogy was all in one.

I was like, oh, I I'm fairly sure I've heard Paolo ranting about how he loves this set collection of books.

So you have to love it, right?

She solves problems with code.

Okay, I'll go.

It's an American book about politics, and uh I'll just say it seems still ever more relevant.

Plus la change.

So uh it's a very good book.

I'm reading it again, and um

Yeah, no, it's a classic.

Anyway, I will refrain refrain from saying more.

Um quick quickly, projects.

I'm gonna highlight um Django Live Translations.

I'll put a link into uh the show notes.

This is a very recent project um in browser translation editing for Django superusers.

So I have st will give it more of a play, but anything that does translations is interesting to me, so I'll put a link in that.

Okay.

Either one of you.

Yeah.

Yeah, we mentioned earlier like Django bolt.

Um that's what that's a project that really got me excited when I read about it.

Um like I said, uh to really use it with La Suite, uh it's quite a rewrite.

Yeah.

But uh maybe with AI we can put it off.

Um I believe he's still taking freelance opportunities, so maybe a special project he can he can do it.

Yeah.

So you know, any listeners who are interested in that could you know help out by just taking a small sponsorship of Fahan there for his work.

I saw that.

My project is related.

Um it's uh it's I'm gonna I want to mention the tech and power benchmarks that have been the the big um web framework speed-off for the last 10, 15, I can't remember how many years they've been around, but they've been around for a long time.

I just wanted to call them out because they're being sunsetted.

And they there was a pull request opened or an announcement opened that they're they're being sunsetted there.

And it's been like, wow, that has been a sort of um sort of pillar in the API frameworks discourse for you know ages and ages and ages, but um yeah, the tech tech empowered benchmarks.

So pour one out for those because that that's been a decade or more of really good work for the community.

Was that new to you?

That that benchmark?

Yeah, it's

I didn't know

I don't, you know, exaggerate this kind of thing.

Um I guess la sorry, very, very last question, Samuel.

Is there any if there's something about Django, if you have a magic wand, one thing you can change, what would you like to change about the framework?

No, sorry, sorry, for WebSockets.

Uh, sockets sockets, yeah.

It depends on you know what you're trying to do or what throughputs, etc.

But yeah, I mean um

And when I the other day said, okay, let's let's use it again, the the engineers who were taking care of the project said, well, no.

I I don't know exactly what what what the problem was, but they said it was not easy to scale and was not optimal, maybe.

I I would I I'm pretty sure.

I'm pretty sure that there's some native, native, more native uh support for this in Django at scale.

Like what we're showing the proof.

Um it's not about channels, it's not about the web the you know, the WebSocket implementation around channels, none of that.

It's about the need to um handle the sync to async jump um in the ORM.

And that's the same everywhere.

So completing the um the ORM's async story is ultimately what's the difference.

Yeah.

Like it's just that that's pushed into the um Psycho PG library rather than into the Django.

There's a nice um but there's a um third-party package now which it is um it's got a kind of proof of concept l stage there for fully async down to the cursor level.

But then as people are exploring it, they're like, ah, but this is happening and this is happening, and those those really hard issues are um coming up.

And this is why it has to be done in a third party.

Um people you know people are always like, why can't we merge it straight into Django?

And the bottom line is because we can't experiment in Django, the stability guarantees and the like the life cycle are too long, and so there's this um.

Um the the yeah, it's coming, but the the the the context managers around which which context you're scrubbing you're sharing the connection between, and do you need to maintain the same connection for multiple calls you know into the ORM over the lifecycle of that single?

It's uh there are so many weird edge cases.

And and the bottom line is there aren't really any async ORMs out there.

There just aren't.

So I think a lot of people that are doing async code at scale, they're they're doing it with raw cursors.

Um rather than

We should find that there would be a huge impact.

No, well, so okay.

When you get it up and running, ping us, and we'll see if we can get an application.

Yeah, yeah.

So yeah, we should be interested to support this committee.

Um and it's not at the view level.

That's all working perfectly fine.

It's that when people need to call to the ORM, they do an A sync to async, and there are complications around that.

So I I have big questions about it.

And and with AI, I I feel may maybe uh you know we we on one way on one side we say we are going to stop writing code and on the other way we we we make the writing the code more painful.

Um like I I in OpenAI SDK I f I found one one day I I I was in a file in the OpenAI SDK where there were literally like 1,000 lines just for the fit function declaration.

Yeah, right.

Okay.

You know, the type the the type of the function it was up to 1,000 lines, and I thought to myself, well, this is going too far.

What what are we doing?

Like I really love the Zen of Python, the readability, the easiness, and like typing for me, uh doesn't I don't I think we made it at least I I'm not using it and I I find it hard to convert to this, so maybe AI.

And AI is not going this way.

It's AI should understand types, should be able to add types without B.

Like that's one of the things where I'm I'm not sure I would like to I I'm not sure where we should go on this.

I gave a talk recently at um JetBrain's Pi TV event in Amsterdam that uh perhaps we'll put the link again in the show notes.

But my talk was um but it was called uh static islands dynamics.

And um the idea is that um Django's core is is designed around dynamic patterns, and so it's gonna be very hard for us to add you know type hints there.

We'd have to rewrite the whole thing basically.

But on top of that, we can put you know nice static islands where they help.

But the lasers is fine.

Yeah, so lasers nice data classes that have and you know the type annotations in there that tell you those are those are useful things.

Yeah.

I think if we try and retrofit typing into the core of Django, we will struggle because there are it's deliberately built around patterns which um Python's type system can't express.

Yeah.

Uh you will

But uh okay, if AI writes it for me, why not?

But I'm I'm going to stop reading Python as well.

Yeah.

I thought, okay, at least one guy agrees with me.

Samuel, thank you so much for taking the time to join us and and thank you for all of your work.

It's very inspiring to see what you and your team have accomplished.

And uh yeah, I I love being in this community.

Uh I hope uh that yeah, there's a lot of nice things coming for our projects.

Cool.

Uh DjangoChat.com and on YouTube, and we'll see everyone next time.

Bye bye.

From scaling applications to simplifying data complexity and unlocking AI outcomes, they help you move forward faster.

See what's possible at sixfeetup.com.