Why are vector databases important at the edge? In this episode, Bill sits down with Dr. Vivien Dollinger, CEO at ObjectBox, to discuss the growing significance of vector databases in the AI tech stack and how this will impact the future of edge computing.
Why are vector databases important at the edge? In this episode, Bill sits down with Dr. Vivien Dollinger, CEO at ObjectBox, to discuss the growing significance of vector databases in the AI tech stack and how this will impact the future of edge computing. They dive into the future of edge AI data management and why more data isn’t yet stored at the edge.
---------
“Vector databases are an essential piece of the AI tech stack and will become even more so.”
“We will see more infrastructure software for edge computing coming up to empower the market to really make the edge as easy as the cloud.”
“I believe there will be more data staying in the edges, residing in edges, than on the cloud. But this is going to take time.”
“This whole terminology around edge computing and edge is not doing edge computing a favor.”
--------
Timestamps:
(06:54) What are vector databases? (excel ata similarity search)
(11:30) What makes it an edge vector database?
(16:43) Data sync and future of data management at the edge
(20:05) What are people doing with the data
(24:30) connections between edge computing and sustainability
(28:30) The future of edge AI data management
(36:00) What’s exciting about edge over the next couple of years?
--------
Over the Edge is brought to you by Dell Technologies to unlock the potential of your infrastructure with edge solutions. From hardware and software to data and operations, across your entire multi-cloud environment, we’re here to help you simplify your edge so you can generate more value. Learn more by visiting dell.com/edge for more information or click on the link in the show notes.
--------
Over the Edge is hosted by Bill Pfeifer, and was created by Matt Trifiro and Ian Faison. Executive producers are Matt Trifiro, Ian Faison, Jon Libbey and Kyle Rusca. The show producer is Erin Stenhouse. The audio engineer is Brian Thomas. Additional production support from Elisabeth Plutko.
--------
Follow Bill on LinkedIn
Connect with Dr. Vivien Dollinger on LinkedIn
[00:00:00]
Narrator: Hello and welcome to Over the Edge. This episode features an interview between Bill Pfeiffer and Dr. Vivian Dollinger, the CEO at Objectbox, an infrastructure software for developers that empowers sustainable edge data management. In this episode, Bill and Vivian discuss the growing importance of vector databases in edge AI tech stacks and how that will impact data management moving forward.
But before we get into it, here's a brief word from our sponsors.
Narrator 2: Over the Edge is brought to you by Dell Technologies to unlock the potential of your infrastructure with Edge solutions. From hardware and software to data and operations, across your entire multi cloud environment, we're here to help you simplify your edge so that you can generate more value.
Learn more by visiting dell. com slash edge for more information or click on the link in the show notes.
Narrator: And now please enjoy this interview between Bill Pfeiffer and Dr. Vivian Dollinger, the CEO at Objectbox. [00:01:00]
Bill: Vivian, so I came across your information and was reading some of your work and it was fascinating and I really wanted to have a longer conversation with you.
So thanks so much for, for joining us on the podcast. I've been looking forward to this for a while. Yeah,
Vivien: thank you for the invitation.
Bill: So just let's start with a little bit of background. How did you get started in technology? How did you get here?
Vivien: Yeah, well, actually, I do think it's all due to my brother, actually.
So he's five years older. So he was rather a tech enthusiast. And because for me, as a Tech gadgets weren't really on my wishlist, but I had all the exposure and the opportunities to play with all that stuff. Like, I'm not sure if you know that in the US, but Fischer Technik, where you can put together light switches and build your own circuits and stuff like this.
And then of course, the Atari came. And that sparked my love for games. And then at some point with the Atari ST, of course, there were all those hacker demos, which I found super [00:02:00] impressive because in my mind, it were just like, yeah, people like me, basically young people, of course, they should have been way older than I was at that time, but still doing something like this, it's just, was amazing.
And so. After school I decided to go for computer science. Love
Bill: it. Okay. So then what made you decide to start ObjectBox? Actually,
Vivien: it all started with the first Android operating system coming out. So, Mark was my co founder, I was super hyped and actually started developing some stuff before the, you know, in the beta, when it was still in beta.
And we just believed This was going to be huge. We also loved that it was open source, of course. And yeah, then Marcus, then basically we built an Android app agency out of this. And at some point, Marcus developed an object relational mapper. That's, you know, something that sits on top of SQLite and it just made it fast and easy to access.[00:03:00]
And because he, of course, open sourced it, it became quite a lot of traction. And at some point. He couldn't optimize it anymore. It was just SQLite was the slowest. piece and a stack. And yeah, well, then we started thinking and thought, why not just get rid of that, well, problem, sort of, and solve it from scratch.
And we believed we could build a faster, easier to use database for embedded devices. And that's what we had in mind when we started ObjectBox.
Bill: Wow. So you really started with the idea of phones, tablets, very lightweight devices.
Vivien: Oh, yes, definitely. So it was, I mean, like by now it's a lot of different devices, controlling units, anything.
It can be anything, but it all started with phones. Yes.
Bill: Wow. So that's just a really interesting thing because as we talk about like IOT devices, they're doing very simple things and just spitting out data [00:04:00] and they can run on a battery for a long time, but actually catching and processing that data, doing something meaningful while running on a battery.
Or while running in such a tightly processor constrained type of environment, that's a whole different set of challenges that you're a very brave woman.
Vivien: Yeah, well, I actually do think that this holds a lot of benefits. Just trying to cloud kind of is the counter paradigm to that, right? Because if need be, you can always just throw resources at any problem.
And it will go away. But if you instead really try to optimize from the bottom up, the performance you gain is actually really sustainable. So it's not just faster, it's also using less battery, for example. So to me, that makes so much sense. If you think this at a global scale, it just is so much more valuable for, you know, just any [00:05:00] economic evaluation, right?
Bill: Wow. Okay. So still thinking about the business, what's your experience been as a co founder? What kind of big surprises did you have? What kind of lessons learned? Would you do it again?
Vivien: Oh, wow. Okay. This is a, this is a hard question. Okay. First of all, so we are a special case again because we are actually family founders.
So Marcus. is not only my co founder, he's also my husband. And when we started Objectbox, we had a very, very young daughter. So balancing a toddler and a fresh startup has, I think, its own, you know, interesting parts and challenges.
Bill: Phones whose batteries run down fast, toddlers whose batteries never run down.
Vivien: Exactly, exactly, exactly that. But I think this is one of the strengths we really also have, because Both of us understand what's going on at home and we both understand what's going on at object [00:06:00] box. And then we can make decisions together. And there have been times where Marcus needed to take care of everything at home.
And I was, for example, fundraising and I needed to close the round. And then there are times where it's, you know, all on me and he needs to finish some feature, for example.
Bill: Wow. I could, I could see that working really, really well. Or really badly. It actually does. Keeping it in the family. I'm glad it's working well for you.
That's fantastic.
Vivien: Yeah, I think it's not for everyone. I agree. It's working for us at this moment. It's quite a
Bill: balancing act, especially with a daughter. So, into the technology. I've only recently come across vector databases. As a technology, I had never come across them before and now I'm starting to see them all over the place.
Maybe because I stumbled across your stuff and I was thinking about it more and it's just, they've been everywhere. Maybe they haven't been everywhere and they're just all of a sudden having a moment. Can you tell me what is a vector database and why am I suddenly [00:07:00] seeing them all over the place? Have they been there and I just missed them?
Or were they there?
Vivien: No, no, no. Yeah, of course. No, no. They had a huge hype this year. And I think it's all due to You know, GPT, chat GPT, all those AI use cases we've been seeing. And of course, AI models. All work vector based, but those vectors actually that us make them so special. Also within the vector database realm is you can actually capture the meaning of, for example, objects, audio files, images as a vector.
And because with a vector, you have a vector space. Having those vectors in a vector space and the distance between those vectors tells you how similar they are. So then you can, of course, do similarity search, for example, and all the semantic search. And this really is what makes those vectors so powerful.
But in order to work with those vectors, [00:08:00] especially if you have lots of data, you need to, you know, store and manage them somehow. And vector data spaces Bases specifically excel at similarity search. They have great algorithms to search within the vector database for similarities. So they empower many of those use cases actually.
And what happened with those big foundational models being released is anyone can create their own embeddings using a foundational model, but then you have those embeddings and you need to do something with them. So you need to store them in a vector database and then you can create any use case.
Bill: I'm trying to get my head around that.
That's, that's, that's a little deeper than I usually go, but I like that. I want to, I want to push a little bit. So that's very cool. What other types of databases do you typically see for AI driven use cases? I mean, if you're not running a vector database, I can see, you know, that's, that's giving you The ability to, I mean, the database itself is tracking the similarities to a certain extent.
[00:09:00] So you can more easily draw conclusions, draw parallels, identify close things. But how else would you do that? I mean, without vector databases, what was there before?
Vivien: That's an interesting question. What was there before? I mean, like vector databases, while they are high, like prominent now, have been around for a while.
Um, there is, I mean, like you don't always need to probably save the vectors. You can also create new vectors every time you want to do it. This is some of those approaches you can also see now, but typically I think that most. AI use cases, especially if it's around similarity search, semantic search, and all of these, they'll be using vector databases really.
So. I do think vector databases are an essential piece of the AI tech stack and will become even more so.
Bill: Okay. So it's [00:10:00] really, they've been there and it just wasn't, maybe, maybe they just weren't really discussed outside of the AI space. And now the AI space has exploded all over, all over. And now we're seeing that more.
Vivien: Well, I think the big thing now is that everyone, you use a foundational model, a large language model, and you just. Give it your documents and create vector embeddings. Just It wasn't something that was really done before, maybe a year ago or something that it all started to happen. And now, I mean, like now with chat GPT 4, even as a user, you can just give it a document and do, you know, create your own AI assistant with it.
Then now you don't need, you have your own vector database and do it all yourself, which developers typically would do, but, but that's the same in small, I would say. Okay. It's just that they have vertically integrated and do this for you.
Bill: Yeah. Okay. It's starting to clear up in [00:11:00] my head now. That's good.
That's good. So you have a Venn diagram that you use that shows an overlay of AI and edge computing and databases. And that center spot that everyone's always looking at, like the sweet spot of where we overlap is. Edge vector databases. So what makes it an edge vector database then?
Vivien: And this is really something, yeah, as I said, really early kind of because we have an edge database.
So we are all around, you know, data persistence on edge devices and data synchronization. And of course with vector databases and AI, everything booming and becoming so we look deeper into the market. It was really interesting for us to understand what's happening there. What types of data are they using?
What are their use cases? And Edge AI actually is already happening partly, of course, but it's also a highly growing field. There are so many cases that will need Edge AI. Now with vector databases becoming an essential piece of the AI tech stack. It [00:12:00] will be an essential piece of the Edge AI tech stack.
At this moment, basically, there are no vector databases for the Edge. And if you think vector databases and you have a look at them, I personally would say it's infeasible that one of them goes and shrinks itself and can run on the Edge. This hasn't happened to any database in the past, and some have tried.
This seems to be difficult for many reasons.
Bill: Yeah. Once you have a big core engine, shrinking that core engine is really tough. Growing it may be easier, but shrinking it.
Vivien: Exactly. So kind of, you need to take something that can run on the edge and build it from there. And of course I'm totally convinced that this will be one of the next things we'll be seeing.
And yeah, as you asked about vector databases, they are highly specific at that point. They need a lot of resources, a lot of requirements, so they're not [00:13:00] Edge capable. They are already, however, also really specific. So traditional databases have started building vector extensions. So for example, like, I don't know, Elasticsearch and many other MongoDB, I think.
Many other databases are either building that or have built that. So they are kind of migrating a little bit into the vector space. There's a huge discussion if you need specific vector databases or if the traditional databases with the vector extension will later be more predominant in the market. I don't know.
I mean, like, Both at this moment have their merits, but I do see vector databases migrating into the more traditional space in order, you know, to deliver the full feature set, which you would typically expect from a database and traditional databases. Of course, migrating to support vector data and similarity search and things like that, which you would [00:14:00] need for the AI use cases and future will show which one will be more successful.
Bill: Yeah. So that, that kind of ties into where I was going next, which is I, I hadn't thought about this much specificity. of a database making so much difference. I was thinking at the edge, you know, you collect your data and you put it into an SQL database or a NoSQL database, whether it's sortable or object based or whatever.
And now we're throwing in vectors and I'm sure there's other kinds. And so do we have multiple databases? Do we have databases that can handle different types? And now we're starting to, like, I can't get my head wrapped around what a What the, we're going to have some SQL type data and some no SQL type data and some vector type data.
Like that's a heck of a database. How do you call all that stuff? But interesting that, that we're looking at potentially just adding vector type extensions. So we'll have hybridized databases maybe, [00:15:00] or I don't
Vivien: know. Yeah. I mean, like I think coming from a traditional database, just like adding a vector extension, it's just data type.
That's not a really big deal. You really need to add similarity search and all, you know, and this metadata filtering and all of this that provides the real value. So it's a little bit more actually that, you know, traditional databases need to add. At this moment, a vector database probably, I would guess is always better, way faster at handling vector data, but therefore it's so specific that yes, as you say, you will probably typically always need a second database for many cases that adds complexity.
And if you think edge, of course, then you. Specifically, you need to think if you really need two databases.
Bill: Right. Bigger footprint, but more efficiency, but more complexity, but faster. Yeah, okay. That's, it's going to be interesting to see, like, what the scale breakpoints are of, do you want to [00:16:00] compromise on the performance and get a hybridized, or do you want to maintain the best possible performance and have two?
That it's just a little bit more complex and takes more, I don't know. It'll be interesting to see where that goes. That's, that's pretty cool. I never thought about that. And it had me thinking about just the mechanics of the data differently. And I really liked that. I would imagine that. Most folks are not thinking about, but what kind of database is it?
And how is that going to change over time? So watching that as another factor that we have to watch over time is kind of cool. So when I looked into ObjectBox a little bit, I saw references to both databases and DataSync. And the data sync also caught my attention because we've got this massive new data that's being created and it's streaming.
So it's constantly moving and you know, you can't miss it and catch up later. You can't batch process it later because it's all pretty much real time. But then identifying. what's useful, what you want to process, what you want to store, what's valuable, [00:17:00] that the data management that's going to come up at the edge is just, I don't think we've even started to figure that out.
How do you see that being handled with your customers, with your forward looking planning? I mean, how are you thinking about dealing with all of that streaming data and identifying what's useful, what's not?
Vivien: Yeah, well, if it's really Something that's like always streaming, then probably oftentimes you wouldn't want to persist everything, right?
Maybe aggregates only. Sometimes there are use cases where people really need, for example, time series data persisted every data point. And then we would support, for example, both use cases, of course. And we have things like async, batching that, you know, is really fast. You have like, can imagine 100, 000 operations in a second.
And so that we can make. Both cases, for example, work, but the decision [00:18:00] which data to keep and which to maybe aggregate at the edge or directly is not our decision. So what we do is we have a developer tool. So the application layer is what gets to decide which data needs to go where and when. However, one of the things that we do is we don't only offer that we persist the data on one device because the edge With one device is not useful, we take care of synchronizing data across edges.
So the developer can say which data he wants to synchronize where, basically wants to synchronize within the connected devices, of course. And we will take care that this data will get there safely. So if the connection breaks, if X devices go down, if whatever, whoever goes down or is not working, or the connection breaks, we'll, once the connection is there, we'll take care that the data [00:19:00] is synchronized where it needs to be synchronized.
And this is bidirectional, and our synchronization server, it can run on a Raspberry Pi. Pi zero. So it's all very low end.
Bill: Super efficient.
Vivien: And I think this is, and this is helpful, I think, because that way, you know, the developers can really focus on the application logic and really can make use of smartly thinking about which data is needed where, and then the rest, you know, we take care of all that pain and hassle that's around sending this data in an, you know, distributed setup around.
Bill: So you're making it easy for people to identify which data is valuable, but they have to make the decision that makes sense. You can't centrally, generically make that decision of, because who knows what the data is going to be. Do you have a sense of people are doing with that data? I mean, that's, it's how they use it.
So I [00:20:00] don't know, maybe you don't see it, but are people Sending it, like, syncing it around to other edge sites just so they have some backup? Are they keeping it on site? Are they sending it to a central place?
Vivien: Okay, that's totally different actually, it's like, because use cases still are so varied, but for example, if you take automotive, and a car is a distributed system, it has several ECUs, like controlling units, an object box can run on such a unit, and it, it indeed also does, and those units want to.
Use data that, you know, they want to use the same data and some of those units may not typically, or didn't have before access to data of those other units, because there's a lot of security stuff involved and running object box on those ECUs. And synchronizing data as needed is something that enables them to have the data they need when they need it.
Just a simple example. And that's all because a [00:21:00] car for many people, I think, is like one device, but of course, it's many devices. And this would be a typical use case, and that's a pure edge case, basically. And then, of course, you can synchronize it all, but the way back up to the cloud. But you, it's partly done too.
We also see, this is a very typical use case, people wanting to synchronize really data on the edge because they want to be sure it's offline available. Take for example, an amusement park where there are cameras taking images of people that, you know, you want to have at the booth where people can maybe print it or buy it, and you also want to give, to give people the opportunity when they're at home to still get that photos maybe.
Yeah, just as an example, so you sync that up to server when there's a good location, when the network is good, when it's there, but primarily you really need to make sure it works on the edge and within the location. Then of course there's some specific data where you also might want to compare different amusement parks, [00:22:00] which is doing.
Well, with what, from a management perspective, so that would be a typical use case and it's always like tiered, but I see the same tiered approach across industries. It would be similar with the industrial setting.
Bill: Syncing data had me thinking about cars. I'm based in Detroit, so that's just, that's always, I'm surrounded by car people.
And so de facto, I end up thinking about cars a lot. I saw an interview with one of the traditional auto manufacturers. about switching over to electric vehicles. And the traditional manufacturers are having a really tough time because they've outsourced so much of their cars and they have a braking system.
And that's, you know, a vendor builds that and a steering system and a vendor builds that, and there's no central. They don't share data across one another, but the electric vehicle makers have this centralized bus that runs through their car and everything's designed from the ground up to use that data [00:23:00] bus and being able to drop in something like this.
to connect into all the different systems and share the data across so that the steering system could connect to the braking system would be kind of amazing. And then maybe they don't have to completely retrofit, redesign the entire car from scratch. That's wow. That's an odd little hole to have my brain fall down, but it was kind of fascinating.
I can see. You know, even not looking across different sites, but syncing data across within the same site is a more interesting challenge and a bigger problem than I previously thought. It's fun because there is so much complexity. When we start talking about edge computing and it sounds, you know, from a marketing perspective, we say, and then you generate value.
And that's so many different specializations and there are so many different moving parts and exposing those and, you know, peeling that back. But what's the actual [00:24:00] problem? Oh, that's a giant problem. How do we solve that? Oh, that's how we solve that. That's not bad. Very cool. You've also had some conversations about sustainability.
And you said somewhere along the way, forcing a decentralized world into a centralized cloud model is wasteful, both economically and ecologically. What connections do you see between edge computing and sustainability? How does edge computing drive a more sustainable future?
Vivien: Okay. So for, for one thing, it's of course, if data is used.
On the edge, it doesn't make a ton of sense to send it around to somewhere, someplace else, to some cloud and then compute there and send it back. This is, it's like ideally you buy food from your local grocery store that's been grown maybe somewhere not too far away instead of shipping it all the way from Australia, right?
Because that's unsustainable. And it's exactly the same if you ship the data across the [00:25:00] world. Even though, you know, you have it already at the place, this is like just illogical. It doesn't make sense. And this is happening a lot. And the bad thing is, and I hope I can just say that here, but what I see happening is huge cloud providers with a lot of, of course, market opportunities.
Let's phrase it that way. They all have an edge now. And the edge is the smallest possible edge you can imagine. And everything else is still sent to the cloud. It doesn't make sense. That's not sustainable. It's wasteful because you waste all those resources along the way. It also, of course, wastes energy on the edge device as well, by the way.
And all that infrastructure we keep up for doing this has quite a footprint. So all in all, this doesn't make sense.
Bill: It also makes a lot of sense from the perspective of thinking about redesigning these things from scratch, right? If, if you've got a database that's designed to run [00:26:00] on a small device, as opposed to taking a large device and trying to jam it into a small device with all the extra features, with all the extra connectivity that you're not going to use, right?
You know, strip the thing down, build it efficiently for the edge. And I would imagine that. Just the efficiency of that operation at scale in and of itself would use significantly less power than sticking it into a general purpose database and trying to sort of brute force it, you know, the general purpose database does a thousand different things you needed to do to run it.
So write something that does those two things and don't have the extra processing power of all those extra features.
Vivien: That's right, too. Just omitting what you don't need. And of course, retrofitting, just using what's already there also makes a ton of sense. Why not? So there's, there's a little more to, to, um, by the edge of sustainable, of course.
And just using the devices and the power of the devices that's available. Instead of using it to wait, it [00:27:00] also pays off, especially now that you see so much data.
Bill: Yes.
Vivien: Like the data explosion and everyone is using it and everyone is just gathering old data. Not everything is always used later, there's a lot of data waste.
And I totally get that people are probably a bit tired. You know, there's like all the sustainability, you need to be careful about everything you do now. And unfortunately, that's true for the digital space as well. And for development.
Bill: And security is going to become a huge thing too, right? We're creating so much new data at the edge.
It's distributed all over the place. We're trying to move it around back and forth to, from clouds, data centers, all these other places. And at some point, there's going to be a security backlash coming on that's going to be interesting. And I wonder what that's going to do to the conversation about computing data locally.
Cause it's. It's a lot more secure, right? Compute it, secure [00:28:00] it, delete what you don't
Vivien: need. A lot more secure. And of course, personally, I'd like to keep my data with me. If there's no need to share it, why would I share all my smart home data with the cloud? Yeah.
Bill: I guess, I guess a privacy and sort of personal security aspect to it as well.
Yeah. Yeah, definitely. So looking forward a little bit, what do you see coming in terms of edge AI, edge data, the way that we handle it? We talked some about vector databases, right? And how that's likely to rise at the edge because it's so much more efficient. But, you know, we've got, I saw the number not too long ago, a couple billion, like 11 billion cameras in the world right now.
I might be off on that, but some ginormously scary number of cameras that are pumping out data all the time and more coming every day. Where do you see all of this [00:29:00] going? You're kind of immersed in that space right now.
Vivien: Yeah. Well, I mean, like it's not going to get less, right? So unfortunately, even though.
I'm not sure if you need all that data. On the other hand, I do think also due to just plain costs, there will be, I mean, like the shift to the edge has been foreseen for ages and it hasn't been happening. So why hasn't that been happening? Even though. It is needed for some use cases. Some use cases just won't work.
It empowers many use cases that people are looking for. And there's so much pain and struggle about not being able to access certain data because accessing edge data sometimes can be still unfortunately and unbelievably really painful. So why is it not happening? Why hasn't it happened despite everyone waiting for it?
And the answer is clear. It's because it's still. Not really easy. And so then if every [00:30:00] project needs to come up with their own solutions from scratch, it's really expensive. It takes a ton of time. It has risks. Maybe you don't have the right developer talent to pull it off. So all those things make Edge projects fail.
And that leads to, you know, all those postponement of distributing. Data in the right way across the edge, from edge to cloud, because it will always have both, right? But at this moment, it's still all going to the cloud, and way too much, and often, even if it's not useful, but the cloud just makes it easy.
So, any project you want to do, a prototype, quickly set it up in the cloud. It's running. Try to do the same thing on the edge, you just face too much struggle and time and pain. And so I think the next thing that needs to happen, it will happen is we will see more infrastructure software for edge computing [00:31:00] coming up to empower the market, to really make the edge as easy as the cloud.
And then I think we will see a huge shift. A huge shift. I believe there will be more data staying in the etchers, residing in etchers, than on the cloud. But this is going to take time. Yeah.
Bill: Yeah. Software and tooling to make it as easy as the cloud. Do you think that's the primary thing that's holding us back?
Vivien: Yes, absolutely. What do you think?
Bill: I think that's a large part of it, for sure. I mean, the physical deployment. One of the challenges I think that we're having is when you're talking about distributing things out to the edge, it's It's harder to touch stuff. And so you really have to understand what's going on.
And a lot of, I think a lot of the data that is created isn't used because people don't know that it's there. Don't realize it, don't understand the potential, you know, and, and you have to know [00:32:00] that much. about how your system is really operating. What data is being created? Where is it being created? How can it be used?
What are the privacy restrictions? Again, you know, once you start unpacking all of the complexity there, it's pretty amazing. It's pretty daunting. And mechanizing some sort of an answer for that becomes, if you can make it push button easy, that would be fantastic. Then a lot of people would definitely use it, right?
Like chat GPT, you know, you make AI that easy to touch. Well, all of a sudden. Everyone can get their heads wrapped around it. They go, wow, that's what AI can do. It's amazing. Now AI has a whole new life and much more momentum. Not because we're all using chat GPT to do defect detection or something, right?
We're asking it, what kind of shoes go with this sort of outfit and what kind of wine goes with dinner and, you know, silly stuff like that, but, but you can get your head wrapped around what it looks like, and I think, I [00:33:00] almost feel like. What's holding the edge back is that the edge is held back. And so people aren't using it and they can't get their head wrapped around what it is, right?
Chicken and egg kind of problem of how do you get started with that?
Vivien: Cool. Thank you. Interesting. Yeah. And then, because now that you mentioned it, I also think that this whole terminology around edge computing and edge is not doing edge computing a favor, if I can say that. That sounds super stupid, but it's, it's problematic.
It's something else for everyone. And Yeah. Honestly, developers don't even really use that term for them. It's something super abstract. I think some business people have come up with. It's not real. Somehow it's not real. Yep. And of course, we always had an edge, but no one called it
Bill: edge. Right. Yeah. But I've had that conversation with a number of people that I'm looking to.
Improve my customer experience. I'm looking to improve my worker safety. I want my car to be safer. None of those things [00:34:00] is edge to me. And if I'm looking to improve my customer experience, that's the center of my business. That's the core of how I make my money. That's not the edge. Well, it's the edge of your IT network.
So it's, it's kind of a marketing term that is probably one of the things that's holding edge back. Yes. So, yeah, I, I don't know. Verticalizing all of that messaging and all of those solutions becomes an interesting challenge. Right. Even with object box, right. Trying to package it up as, this is a retail solution to solve this thing.
This is a manufacturing solution to solve this thing. You know, how much, how many marketing dollars do you have? How many marketing people do you have? Can you speak all those languages? The retail lingo and the manufacturing lingo. And, you know, they all use different terminology and are trying to solve different problems.
It's It's, it's crazy stuff. So from your perspective, how would you attack that problem? I [00:35:00] mean, I presume you are trying to attack that problem some. Yeah.
Vivien: Yes. Kind of. Okay. So like we're a small startup, right? We're a small team. Yeah. And so what we are doing with object box, actually, what's great about it is you can use it anywhere.
It's like, you know, it's a database for data synchronization. You can visit a mobile phones in an amusement park within the car and on the shop floor. That's great. Right. And this is also. The biggest pain point we face as a small company, because that doesn't work. It just doesn't. And yeah, everyone told us, of course, again and again.
And we were like, but we are already so focused because it could do so much more. And you know, we, we, we became more and more and more and more focused. And at this moment, we are basically looking at very specific cases from our side, but we We actually do get regular inbound leads from all over the place.
And we of course take those calls and [00:36:00] we see if we can make that happen. And this gives us exposure and learning of, yeah, new language and new use cases that are not as specific as the ones we picked.
Bill: I love it. So you picked a couple and you're focused on them, but you can still sell to whatever.
Because you could do it. That's brilliant. Yes. It's brilliant. It lets people understand how it can solve their problem. And then if they have a different problem, they can ask about it. That makes a lot more sense. So jumping back toward the future, what excites you the most about where edge computing, where object box, where vector databases are going over the next couple of years or are likely to go over the next couple of
Vivien: years?
So I'm absolutely convinced that every database will have a vector extension that will be a, basically like a commodity. That's just what I'm thinking. And as I said, as I believe vector databases will be a central piece of the [00:37:00] typical AI tech stack. And so I see that happening. On the edge as well, and for me specifically, I do love distributed systems and I like making use of the power of a distributed system.
And now, of course, with Edge AI, you need to take care of the distribution of model updates of, you know, exchanging parameters and stuff like this. So this is what excites me most because this kind of, it combines everything. We've been interested in doing up to now to something that I think can be really powerful.
Bill: So good. How can people find you online and keep up with your latest work?
Vivien: Okay, I think so. It's just, yeah, Objectbox. You'll find us on LinkedIn, Twitter. We have a website, Objectbox. io. Of course, we are on GitHub.
Bill: Sounds good. Vivian, this was wonderful. I learned a lot. I had fun. I hope you did too. Thank you so much.
Vivien: Thank you. No, it was great fun. Thanks for [00:38:00] having me.
Narrator 2: That does it for this episode of Over the Edge. If you're enjoying the show, please leave a rating and a review, and tell a friend. Over the Edge is made possible through the generous sponsorship of our partners at Dell Technologies. Simplify your Edge so you can generate more value.
Learn more by visiting dell. com slash edge.