340: kerollmops — From Hackathon to Success: The Meilisearch Story
Download MP3I don't see a real contradiction between both sides. Like when we please the open source community, we also please the customers.
Arvid:Hey, I'm Arvid, and you're listening to the Bootstrap founder. Today, I'm talking to Caro, the technical brain behind the open source search engine, Maili search. I've been using this blazing fast tech for my own business, PodScan, and Kero has helped me through a few challenges over the last weeks. I really wanna dive into the economics of building a business on top of an open source piece of software. So let's dive right into chatting about self hosted search engines today.
Arvid:Kero, first off, thank you so much for being present on Twitter. It's it was really nice over the last couple weeks months to just talk to you all the time about technology. It's really cool. And I think rarely do developers, not just me, but a lot of people get to interact with the people behind the tech that runs their infrastructure. So this is really cool.
Arvid:Let's talk search today. I'm really excited. I think VeriSearch 1.9 just came out a couple weeks ago, and it has a lot of vector search features. And I wanna ask you because you're the one building it. Is this the future of search, like similarity search, retrieval, augmented generation?
Arvid:What's vector search, and how does it change what search is all about?
kero:Oh, yeah. So, yeah, I'm very happy to help you on Twitter because you are one of the biggest kind of customer we have with the most documents and the biggest documents at least. I recently, like, dived into your sometimes and and that I you you gave me and, woah, your documents are very big. So, yeah, I will have to to work on that. But, so, yeah, hybrid search is kind of, it's not kind of, it's a mix of keywords search that you already know for sure on a lot of websites.
kero:And semantic search, which is like you probably already know, based on embeddings. So it's like, move the document in a space and search for when you search for something, search for, documents that are the nearest to your search point. And, the great advantage of, the semantic, new semantic system is that you it understands semantic meaning. So for example, if you are searching for some things that you can't express with keywords because you don't know what you are searching for, you will find it anyway or at least something that is near what you were searching were you searching for. And that's where many search is very, very good.
kero:In terms of keyword search, we are already very, very relevant. We have already done a lot of benchmarks. And we also were inspired by Algolia, but I will probably talk about that later. Algolia is very good in terms of keyword search. Unfortunately, they don't support semantic search.
kero:So Medisearch will mix the score of semantic search. So it will take you to query and mix the results with the keyword search by using a score. It's not a fusion ranking. It's much better. We will not dive into that.
kero:But anyway, at some point, we will raise real benchmarks about how many search does that. And if your semantic results are not very good or very bad and the keyword search is very is better, those keyword search results will be brought up very high in the results. So
Arvid:Oh, that's cool.
kero:And by the way, yeah, by the way, the structures that we use inside of Medisearch was created from scratch by Tambo and myself. It was highly inspired by the Spotify Annoy library, which is a C plus plus library. It was one of the reason why we did that is that the library from Spotify is using a lot of memory, memory mapping, and Medisearch is also using that thanks to LMD. And we were thinking that it would be great to put all the data structures that annoyed was storing on disk, put that into LMDB, so that we are supporting ST transaction and whatever you want, whatever Medisearch is already using and and and works well for search speed, and we succeeded. There are 3 bullet posts on my on my blog, and you can find them on the Internet.
kero:It's very deep, I mean, tech. But at the end, we are I will summarize that, but we are very, very fast faster than the Android library. We use less memory, and we are able to afford to, to to index, something like 15,000,000 non quantized, vectors into Medisearch in, like, an hour something 1 hour, something like that, which is very good.
Arvid:That's crazy. Those numbers those numbers blow my mind. Like, every time I think about just a massive amounts of data that these these engines can take in and then meaningfully put somewhere so you can then very quickly access them. Just crazy. What what it's supposed to funny to hear you talk about this, to hear you talk about the Noy library and LMDB, both of which are effectively open source projects too.
Arvid:Right? Like, it's like Maillesearch as well. It's like a an open source project. This is this is really cool to see. I I actually dove into the of the issues and mainly searched recently because I was I was trying to, you know, just look at how it could improve my own installation, and I just looked into the issues and how you guys on the other side of the fence in the development team deal with little problems that you might encounter.
Arvid:And I saw you in the bug tracker for LMDB talking to them about some kind of memory issue in their own thing. It is so cool to see the open source community that that consists of these different little villages, each of which is a project, but all of them are kinda trading with each other. All of of them are interacting. So it would be very interesting, I guess, to hear about how how ManySearch started because it is an open source project, so everybody can use it. But there also is a business side to it, which is super interesting to me.
Arvid:So can you explain to me how this all happened? Both of these things, I guess.
kero:Oh, yeah, for sure. So they are very related because, before starting coding on Melisearch, which was called Raptor at the beginning. It was, I was, sorry, a very big open source contributor, and I'm still was I am still is. I'm I contributed to the Rust Standard Library a lot. Like like there is maybe 4 of my, contributions on used by a lot of people in Rust today.
kero:I also created a lot of libraries and stuff like that in open source and contributed to a lot of other libraries. And so, Medisoc started when I was graduated from 42 school, the 42 school, French free school. And I started working in a French e commerce platform, and with Quentin and Thomas at some point. I mean, the quality of the search engine of this company was not very good. For example, when you were searching for a red dress, the engine was giving you red wine.
kero:What the heck? So you are an e commerce company and you are not able to answer your customer, requests. So what the heck? It is one of the most used feature of the of the of this website. So it it it has to be it had to be fixed.
kero:So we started, looking at the Algolia's blog post about the way it works and stuff like that. But I was, at this point, hired by the company and no longer working on this specific subject. But I was very interested by it because they were working in Golan. I was like, no, it's not a good thing because there is a garbage collector and stuff like that. So I was like, you should go to Rust.
kero:And you were like, it's too hard to cut in Rust. And I'm like, yeah, I know, but you will not do C plus plus anyway. So at the end, they decided to go, with GoDaddy. And I decided on my weekend to cut on my own search engine from scratch. And something like a month later, mine was way better, more tested, faster, whatever you want, working on RocksDB, and called Raptor back then.
kero:And both of my friends told me, oh, do you want to create a company around that? Because we were it's boring there, where we are. Let's go create something. And we are we were so lucky that when we came back to school because we we we needed some place to work, hosted a hackathon, about creating the future of search around their internal catalog for their sales assistants working in their retail stores. And so we crafted the search engine features that they needed.
kero:And at the end, short story, we won against Algolia Analytic Search and we worked for them for 9 months. And, so I will go on these details, but it was very important for us to, keep the engine, open source. And they absolutely were very, I mean, happy about that. And they decided like we will sign a contract to make sure that it keeps that way. Since then, we raised 3 series in 4 years and we are now 25 people in the company.
kero:Wow.
Arvid:That's awesome.
kero:Yeah. Fun fact, on the e commerce website, French e commerce website, now, so 5 years later when you search for a white shirt, you find white knives. I mean, knives with a white hand. So, yeah, thanks thanks for not using medicine. So yeah.
Arvid:That's funny. Oh, my. What a story. I I love this. I love this.
Arvid:And and, again, like, this this is, I think, like, probably the 100th time I've heard somebody who had a problem and needed a solution, and somebody else didn't really want it. They didn't really understand it. And so, yeah, they build it themselves, and it turned into a business that is actually working well. It's so cool to hear where this came from. Out of frustration of the the whole red wine, red dress situation, that is really, really fun.
Arvid:Cool. And also interesting with the hackathon, like, bringing bringing these, bringing people together around a commonly common interest. That that is just the best way, particularly in open source to get to move anywhere. Right? Like, how many how many contributors core contributors do you think, Maybesearch has at this point?
kero:Oh, I mean, it's very restrained because it's a search engine at the end at the old. Sorry. And it's like something like, I would say, we are in the company's biggest contributors. But Medisearch is not only one single repository. As there is a million inside but it's in the same repository.
kero:It's the greatest thing that provides a search for a single index. And we have a lot of dependencies. So for example, roaring bitmaps to do a lot of operations on sets, integer sets, and also, HID, which which is maintained by Medisearch, which is a wrapper around Lmdb type, wrapper for Rust, around Lmdb and a lot of other stuff. We have more contributors on the smaller smallest libraries like HID or, sorry, roaring bitmaps and stuff like that. On Medisearch, we have a lot of people coming in to fix small things like for example, when Medisearch do not have enough space on this, for example, it needs to grow.
kero:Before, like, a month ago, it was growing by 10 megabytes megabytes, which is very not much. And someone just helped us on that and make it be 1 terabyte. So whatever you don't whenever you don't have enough space on disk, it grows by 1 terabyte, which is a quick, easy fix kind of. So most of the time, it's not big complex contributions that you see on the search engine. It's too complex for someone that doesn't work every day.
kero:But if you want numbers, I will say, like, a total number of people working on research could probably be around a 100 a 100. But for sure, like, we have maybe 5 contributors, on the wall SDKs that we have. So for example, Ruby, PHP, we have a lot of people working for ManySearch, which is not the main core engine, like SDKs mostly.
Arvid:Okay. Well, thank you. Those are still interesting numbers. Like, most most people who start their own the open source projects, they build everything themselves, and then there's maybe 1 or 2 other people or the users come in and they report issues, but they don't really contribute. Right?
Arvid:There's no pull request. There's just complaints. So it's it's interesting to see the size of the operation. And operation maybe the the interesting part here, because there is the open source project, and there's all the SDKs and all these little subprojects and adjacent projects. I guess there's a website too that needs to be updated.
Arvid:Right? There's the the maybe search the PHP library. There is the, there's a JavaScript library. And all of these things, they need to be maintained. And then there's a business side of things where you you don't, you don't just excel software.
Arvid:You sell hosting. You sell, like, operations. Right? How do you split your time between contributing to stuff that goes purely public and things that are internal that you need to build?
kero:So yeah. You're right. The main things that many such, the cloud offering is selling is like an infrastructure, like more than the search engine. You can use a search engine whenever you want. But the the I mean, making sure that it stays up and works well, doesn't crash, whatever is maintained by the cloud.
kero:Like, we we are better than most of you to understand how the engine works. But yeah. I mean, we can help, but, you know, it's it's a world job. So, yeah, what we are really selling I mean, we have a team that works on, the AWS infrastructure to make sure that the engine works well. And whenever we have a big customer like you, with more than 10 of 1,000,000 documents, we are we are making sure that we we select the right, machine, and the we we we gather the logs.
kero:We analyze them and whatever we we need to do. I don't often work on the Kubernetes platform we have. I'm more on the engine side. I'm very like a tech guy who works for, with LMTB a lot, with, making sure that we when we read something on the disk, it's fast, quick. It's not I mean, it's it's in the OS cache, thanks to memory mapping or whatever you want.
kero:I'm very deepened into that and not a lot on the Kubernetes settings. But we we we we have a team for that, we dedicated to that, the love, like, Yamel. I don't know how.
Arvid:You're right. I don't I don't understand that either.
kero:Yeah. I don't understand why they they like that. But, yeah, they they do that much better and more than me. So I'm, I'm kind to have them, in the team.
Arvid:Yeah. Yeah. And it's it's I love this. I love that there's people who really, really love Kubernetes. I don't know I don't know what what's wrong with them, but they like it.
Arvid:And and the thing is perfectly fine. Like, when people like to to build these infrastructures and I I personally know a lot of people who love this kind of stuff. Right? They they don't wanna code. They wanna configure.
Arvid:They wanna set things up. They're they're builders. Right? They're not they're not makers. They're builders, which slight difference.
Arvid:And I, either way, they're very much needed to to build these stable systems and keep them running. So good good for you to have found these wonderful people. Right? That's just really nice. And, it's also cool to see the the effort you put into making sure these things run-in a stable way.
Arvid:Because if you if you are a serious business, you need to have this critical part of your infrastructure. You described it with the the whole red Ryan, red trust problem. Like, that is, if if not the most important part of any ecommerce site or any any search based product. And my my own thing, like, PodScan is very much search based. Right?
Arvid:My API is search based, and my interface is search it's because it's it sounds like you're you're aiming, maybe search to be the tool for big companies as well. Like honestly, I I when I saw it for the first time, I thought, okay. They wanna replace Elasticsearch. They wanna be the biggest player in the field. Is that right?
kero:Yes. It's kind of that. I mean, what we want is that we want to, make sure that ecommerce companies stop using Elasticsearch because it's a pain in the ass to set up. You don't even have real, type utterance. You need to use n grams and whatever you want, a lot of plugins that probably works well, but I mean, it's so hard to maintain that you don't want to do that.
kero:But what we wanted to do is like the open source the open source, solutions that you can use in a click like just put your credit card, click on this button, and every day, you will crawl your your documents, your new new documents updates, and whatever in in too many search. And it works, very well, since the beginning.
Arvid:Yeah. Simplicity. Right? You're looking for for making it really, really simple. I like this.
kero:Yep.
Arvid:Well, it attracted me. Right? I I can tell you maybe the story how I found my way to ManySearch. Well, maybe that might be just just interesting. I I built a Laravel project.
Arvid:I'm I'm, rekindling my PHP knowledge. Like, I think the last time I built anything in PHP was 2,005, and then I stopped. Then I went into the JavaScript world, and I went into the Erlang and Elixir world into functional programming, which is very exciting and a lot of fun. It was better than PHP at the time, but I came back to it last year, and I found PHP to be very different. The ecosystem is massively better than it was back in 2005 when I left it.
Arvid:And I found, Laravel, the the framework, and Laravel scouts their search engine implementation. And I think what Taylor Otwell did and and all his many colleagues in the in the Laravel space was picking a couple of really good solutions for Scout to support. Right? And one of them was MailiSearch. And I was like, oh, what's this?
Arvid:This sounds fun. Right? I looked at it. I looked at the website, which is wonderful, by the way. Like, both, the website itself is beautiful.
Arvid:The API documentation is spot on. It's really useful. The guides are really good that guide you through the process of setting it up or if it just figuring out how to use filters, how to use sort of rules. All these things that make sense after you get it. But before that are kinda complicated, like any DSL talking about Elasticsearch.
Arvid:Right? The world's worst DSL to describe a query. So Elasticsearch is really, really hard. Maybe search is much easier in that regard. So it attracted me to the project, and then I installed it, and it worked immediately.
Arvid:So, yeah, of course, I'm gonna keep using this. Right? So that that is my story of how I came to use it.
kero:It's very funny because, I mean, you say that Mediaset was funny. I mean, it attracts you. And the fact is that Meilysearch, the name, Maille comes from the, Norse God of oblivion. It's kind of the same as Discord. It fixes Discord by and they gave them this name.
kero:And we found out later that Mei Li also means beautiful in in Chinese. So it's quite fun too. Yeah. And you you are talking about the documentation of Mei Li search and the fact that it's very good in the guides and and all the filtering and stuff like that, you will be you will make someone very happy. I mean, Guy Guy is doing a great job, in in the Medisearch team to to maintain the documentation.
kero:And we are we are also working hard on making the website faster, because you probably noticed some slowdown. But yeah. Yeah. We are working on that very hard. It's very hard to to maintain a good website.
kero:I mean, a quick website, but yeah. And also also a fun thing. I mean, Laravel was PHP. Laravel was my first language ever. I started with that when I graduated by my baccalaureate.
kero:And first, I mean, when I was working in Laurel, I first saw that it was a language. In fact, it's PHP, but I wasn't good enough in computing to understand that. Yeah.
Arvid:Well, the and that's the thing. You learn these things over time, but it I think what Laurel does so well is just be easy to use. Right? You might just well think it's a language because it's just okay. Like, I this is it's well documented.
Arvid:Everybody, like, knows it right through its, like, popular distribution. There are a lot of people that have written tutorials on it. There is a lot of help. There are a lot of plugins. The ecosystem is massive.
Arvid:You don't. That's the thing. You don't even need to leave the official ecosystem of VeriWell to set up a fully functional web app. Right? You don't even need to install another library.
Arvid:It all comes from the official libraries, which is which is really fun. It is it is so funny that the thing you just said about the website being slightly slow, it's, I mean, normal problem for anybody who built is building website, but it's just so funny that many search is so fast, yet the websites that are not built in Rust, I mean, not completely, I guess. And unless you use Firefox, which is kind of partially built on Rust, I guess. Right? Like, all of all of this browsing and the the the interpretation of JavaScript slows the things down so much that the website about such a fast project is slightly slower.
Arvid:That's just funny to me from a technology side. The choice of Rust is interesting to me too. Like, I think, I I'm not very well versed with, like, low level languages. That's what I would call it. Right?
Arvid:I I I I work on interpreted languages or things that are compiled like Erlang or or Elixir, but Rust to me feels anything that touches memory scares me. Let's just say that. If there's a is there an allocation of memory, I'm not gonna touch it. So how how and why did you pick that particular language? How did you get it to Rust?
kero:Yeah. So it started in, in the school, so for this school. Yeah. I I I learned, c plus plus back then. And at one point, I decided that I needed a language for me.
kero:I mean, a language where I was better than the other students and stuff like that. Was very good in c, but, you know, it's c. So you don't as you as you said just before, you don't want to allocate something in c because you will forget about it. Or you will try to free it and use it just after c. So it's like, I need something that is more safe and as fast as C plus plus So I decided to try Rust, on the v one dot zero, the announcement, something like around the announcement day.
kero:Yeah. Yeah. I I decided to start with that. The the compiler was crashing in every day in every way. You you can find, like, trying to compile something was just not working.
kero:You are I mean, it was so hard to work with lifetimes. But and and therefore, I decided to go back to c plus plus because I was able to combine in something. I mean, to work at at the end. And after a week, I decided to come back to Rust and force myself because it was so hard to just have a constructors and a named constructors in OR. C plus plus was so so old and complex to use.
kero:So yeah. And after that, I took, like, a year to do again all of the projects that I did in the last 3 years in the school in Rust just to make to make sure that I I was able to compile programs, any programs that I wanted to compile, any programs that I wanted to to see go faster and be more efficient and whatever. So I built, like retrosell, replicated on multiple machines, a lot of, the worst sort of puzzle of, I mean, yeah, programs that that were working in Rust very well. And yeah, that's that's where it goes. And that's where I started to work on the standard library of Rust from from time to time, finding stuff that I needed that way or not in the standard.
kero:And I also tried Haskell. Not Helen. Yeah. But Haskell. It was slow, very hard to understand, when you have a compilation error, a compilation message.
kero:It's cool, though, because it it created Rust kind of with OCaml. And I don't know if you know, but the first version of the compile Rust compiler was in OCaml. So when you are compiling, I mean, when you are using Rust, which you are not, it runs OCaml in the back, at some point.
Arvid:Well, I kinda do. I mean, I I I built my own Mailysearch binary. So, technically, I I there's a Rust component to this.
kero:No one told me that it was very easy to compile. Yeah.
Arvid:It was. I I was surprised. Like, I I never touched a language before, but the tooling, like, has has gotten really good, must have. Like, cargo and all that, that is really, really stable. And I only needed to change 1 or 2 files for it to work, which, in in my experience, compiling things on a on a lower level never happens.
Arvid:So that was really cool. Yeah. Really appreciate your guidance. And maybe maybe it just really I I set this on the beginning of this conversation. I really like just how present you are in the community.
Arvid:With me in particular, because I got to benefit from this, but I see you talking to a lot of people who have issues, who have questions, who have just an interest in in Mavysearch, in the things surrounding it. And you're always there, and you're always helping them out. I think this is so cool just from a from a perspective of somebody who loves software to see somebody helping somebody like you, helping other people get things done. It's really, really nice. Really appreciate that.
kero:Yeah. Yeah. I think that's one of the things that built Medisearch, I mean, a lot. Like every time there is someone who created an issue on Medisearch, we took less for sure, less than a week. It seems maybe a lot, but it's very not, a week to answer someone and to make sure that this person is using his engine correctly.
kero:And if if if yes, if if it is, we want to understand what you did just to fix it the quick quickest way possible. So, yeah, we are doing that a lot. Also, we have a Discord community with a lot of members that are helping them, helping each other a lot. So we are kind of interacting with them but less and less because people are knowing Medisearch more and more. Medisearch is well is more and more documented, well documented.
kero:And people find the answers to other people just because they already had this kind of issue or this issue was has been fixed in newer versions of MiniSearch.
Arvid:Yeah. That's well, that's great.
kero:One of the issues in is in the indexing speed, for sure. We improved that so much. We are doing benchmarks that we'll probably have, next week. And we are working I mean, I am mostly working on your issue, of it into and mixing a lot a lot of documents.
Arvid:Yeah. Big, big documents. It's so cool. I I'm honestly, I'm glad that I get to share some some part of my database with you just to see what the real use case is that I have. Because I think that's that's probably one of the hardest things to come by if you have an open source project that is used by so many people is to see, like, all the little edge cases.
Arvid:Right? It's to you you get reported on edge cases, but you don't really see the data in them. So it's it's nice to have a community like this where people share with you what they work on.
kero:Yeah. One of our biggest customer is, like, 225 sorry, 276,000,000 documents, so more than a quarter of 1,000,000,000 documents updated, like every 6 hour. I think 100 sorry, 1,000,000 of documents are updated every 6 hour. And you are coming with your 2,000,000? I think it's something like 2,000,000 documents.
Arvid:Yeah. We'll get, like, 2.4 or 2.5. Yeah.
kero:But your documents are so big. That's that's the main issue. I was like, but we are, I don't know, to to to manage 2,000,000 documents. And then I saw the size of them, and that's the main issue. Yeah.
kero:So we are working well on that. And we also have some funny I mean, customers who want to update documents in less than half a second. So we are like, okay. What's your real use case? Because we can help you But updating your wall database, in half a second is probably not possible in any, database you can find on the Internet.
kero:Maybe. Yeah. Or you keep that in RAM, but it's probably not possible.
Arvid:That's like very close to real time Yeah. Like updates. And and that is that is the dream. Right? Like, if you're if you think about the the whole, new asset as a concept and all of that.
Arvid:So, like, real time is scary in many ways. Right? Because it costs you a lot. Cool. This is so nice.
Arvid:It's so nice to hear all the different use cases. That means you must attract quite a lot of different companies. Can you maybe talk a little bit about, like, who is using Maillesearch? Because, I mean, I am with PodScan clearly. It's and it's great, but there there probably are a couple more companies somewhere out there using this.
Arvid:Right?
kero:Yeah. Yeah. Fun fun that you asked that because, last, months, I was in New York, with the founders. And we we meet Arch from the company. I have a cap of, the company there.
kero:Yeah. I'm very proud of that because it's very hard to to to talk to him. And he is very a big fan of Medisearch. Like, he found Medisearch in a very deep repository interest somewhere. I had never seen this repo before.
kero:And they were they were mentioning research. And I'm pretty sure there was a typo in the name because because it's very hard to to write research, I know. And we also met I met, Tinybird because they are a friend of us. We work with a lot of companies. Louis Vuitton is still using a version of research that is not versioned because it's so so hard that, yeah, they they are still using that and it works, well.
Arvid:Yeah. That's also funny. Like, keeping keeping people updated on this or just I mean, if the non versioned version of Mailysort is already performant enough for them to use it, then you're a really good coder. That's really nice to hear. Wait.
Arvid:Where do you see this going? Like, do you because obviously, there there are 2 things that could go somewhere. Right? There's the the open source project itself. That will change, and we talked a little bit about vector search, hybrid search in the beginning.
Arvid:There's a lot of AI and embeddings and things that people want. The people want real time updates. People want speedy indexing, there's a lot of stuff. And then there's the business of the the major such cloud as well. Like, where where do these things go?
Arvid:Are they very aligned? Or do do does the open source community have different ideas where it should go, but the business has that one? What what is the position there?
kero:Yeah. A lot of maybe you you may know, but a lot of people comes from the, open source side. Like, they they use many such on their own. We have seen a lot of, even CTOs of companies that comes to many such as customers because they tried Mediaset on their own. And what's cool is that most of the ideas from the open source users, goes also, works well also for the customers.
kero:Like, for example, the the a lot of customers want better search. And therefore, they want vector search and hybrid search and whatever you want around that. They are seeking for high availability, which we provide with the cloud. As we boot very well and stuff like that. We are about to also replicate your content on different machines.
kero:And, that's what the I mean, a lot of our customers also support only customers. Like like, for example, for you, imagine, of it. Like, we we provide support to to people who are using Mediaset, needs help, and, don't want to share the content on our cloud because it's a security reason. We can understand that. And for you so a lot of, subjects are on both sides.
kero:And I don't see a real contradiction between both sides. Like when we please the open source community, we also please the customers. There is maybe a difference between the the embeddings they are using. A lot of open source people are using a Big Face, the open source, embedding systems. And a lot of our users on the cloud are using OpenAI.
kero:Yeah. Which is
Arvid:Yeah. I mean, that's obvious. Interesting.
kero:Yeah. Yeah. This.
Arvid:Right? Like, it's like they're already using cloud stuff. They they're they don't mind using another API. Well, I'm I'm also on the other side. I'm on the kinda I want my things to be local.
Arvid:I wanna have a local BERT and better or something something that I can can control or I know that it's always gonna be there. Right? You know, the the the whole problem with OpenAI's cloud, like, sometimes it may be down or they may change a version that you don't want or whatever. That's it's always a little issue that I think lots of open source people probably understand, like, the complexity of these dependencies. But, yeah, what I hear you what I hear you say is you're you're working on making the core features of search better and more reliable, faster, and more exact, more precise.
Arvid:Of course, this benefits everybody who's using it. Right? This this is so nice. It's nice to have, like, one core product that you can just improve, improve, improve, and that also helps the business.
kero:Yes. Ex exactly. Like, if you try I mean, most of the time, people, as I said, are trying Medisearch on their own before going to the cloud. And if you please, people, on the OSS side, they will, rather talk about you or also start to to to want to use you for other businesses. We have a lot of customers that are, for example, stopping their company or whatever.
kero:And let's say, we will still be using Medisearch for our next company for sure. It's it works so well.
Arvid:That's a that's a great way of marketing, I feel. I think this is a Yeah. People market themselves into your business. Like, they they are your own like, you don't need salespeople. They sell themselves.
Arvid:So it's kinda how that works.
kero:We still need we still need salespeople.
Arvid:I I I hope I hope you do because otherwise, it would be very risky to to just leave that to people. Was there ever a time where it was really, really hard to get customers into your cloud offering? Or is that still hard? Like, how how has that changed over time?
kero:You know that the first version of the cloud was, way at the start of of many search. I mean, in 2014, we had a first version of the cloud. The fun thing is that there was no documentation on how to use ManySearch. I mean, nobody could even guess what the API of the engine was. And we were like, why don't we see anybody using the cloud?
kero:It's just because there is no documentation. It was obvious. So then we were short on on money, and we did our first, founding. And, we started to understand that, yeah, people can use MiniSearch if they know how to. So we started to use s to to raise SDKs for PHP, I think, JavaScript first and and stuff like that.
kero:And, yeah, on on the number of people that are using the cloud, I mean, it's increasing every day a lot. And the thing is that recently, I mean, maybe a year ago, we started to advertise more on the cloud because we people didn't understand that we are the cloud because we were open source. And it's not something that is very obvious. We only know, I mean, a few companies where we are we are doing that. So for example, Strapi, maybe, you know, is our friend's office.
kero:They are French and they are friends office, and they are doing the same. I mean, it's open source. You can install it. Whatever. I will not, for example, because I don't know how to cut in JavaScript and in whatever you want.
kero:But I I prefer to use the the cloud offering. And we are doing I mean, Zip probably paved the way when we started, say, we're already there. But it didn't help much, because it's not exactly the same solution. But, yeah, we are trying to advertise more, on documentation and wherever we can just so people know about the fact that we have a cloud. And you don't want to to have a burden of set up in many such yourself.
Arvid:I think the strategy that you have in in giving people oh, like, it I felt I felt this too. Right? That when when we first talked, you you just helped me. Like, you didn't expect me to buy anything. You didn't expect me to, like, migrate everything immediately.
Arvid:You just helped me. And now that you've helped me for so much and so long with all these tiny little things along the way, like, obviously, it's much easier for me to consider moving everything into the cloud. Right? Like, this is clearly a good strategy that works for me. And if it works for me, it probably works for a lot of other people in the open source community too.
Arvid:It's so funny because I always tell people who are building tools for other developers that they are one of the hardest people to sell to because developers love a challenge. They love building things themselves. They don't wanna pay money for something that they could do themselves. So it is an interesting approach to just be very free, be very, noncommittal, but still be present in and and communicate with them and then get them into your ecosystem.
kero:Yeah. It makes me think about the fact that Medisearch is very, like, the culture of Medisearch is very diverse and very, we are trying to like, for example, the the leadership team is half woman. And we are trying to be very, like, kind. Like, if you want to use a product and pay for it, go on. If you don't, that's not an issue.
kero:We'll help us in any way by reporting bugs, whatever you want. Yeah. Because we need such add a bug once. But, yeah, we are trying our best on that, to, like, yeah, be kind and let people use message if they want, pay for message if they want to.
Arvid:That is really nice. I love this. This is very empowering. I think you're giving a lot of people a lot of opportunity that may otherwise never have thought about using the product or using the platform. Right?
Arvid:Like, it's it's really, really cool. It's very embracing, a very community centric approach. And the fact that you have a Discord where people can come and talk to you, that you are so present on social media, All of this is to me a very strong, yeah, signal of culture. That's exactly what it is. A culture of, empowerment and of of being positive with the community.
Arvid:This is really cool. So well, if people wanna know more about you, if people wanna find you in particular because you're really cool, but also maybe search and the the cloud and all of these things, where do you want them to go?
kero:Oh, yeah. If they want to find me, they should, reach me on on Twitter, Keyuromaps. Keyuromaps, you concatenate the buzz of these words and you will find me on Twitter. And about Medisearch, I it would be great if people can register on medisearch.com/cloud if you want or medisearch.com.
Arvid:I I highly recommend looking into Medisearch as a as a I have a I have a testimonial on my website for PodScan that is just somebody saying how cool they think it is, how fast the search works. Like, they don't talk about anything else. They just talk about the search. So that's you. Right?
Arvid:That's you and your product that is doing this. And if if anybody wants this on their own SaaS and their own business, go check out maybe search. It's really, really cool. I'm not paid to do this. I'm I really like it.
Arvid:I really appreciate it. And I I it has been one of the most wonderful additions to my software stack, in the recent years, and I will not shut up about it. Thank you so much, Karel, for being on today. It was a really nice conversation. Thanks for nerding out with me about search.
kero:Thank you, Avid. Thanks very much.
Arvid:And that's it for today. Thank you for listening to the Bootstrap Founder. You can find me on Twitter at abidkahl, a r v I d k a h l, and you'll find my books and my Twitter course there too. If you wanna support me and the show, please tell everyone you know about PodScan.fm and leave a rating and a review by going to rate this podcast.com/founder. It makes a massive difference if you show up there because then the podcast will show up in other people's feeds.
Arvid:Any of this will really help the show, so thank you so much for listening. Have a wonderful day and bye bye.
Creators and Guests
