Building computer vision tooling with Niko from Rerun

Niko West 0:00
for these products, like the new knowledge about how the product works is always generated by the developer working on the product. They really understand that deeply. So that's where the real new knowledge actually is generated.

Jack Bridger 0:14
Hi, everyone, you're listening to scaling dev tools, the show that investigates how dev tools go from zero to one. I'm joined today by Nico, who is the co founder and CEO of reran Nico, thanks so much for joining.

Niko West 0:27
Thanks. Thanks for having me. Excited to be here.

Jack Bridger 0:30
Yeah, cuz I know you're working on some really, really amazing stuff with computer vision. are helping companies that building with computer vision, can you tell us a bit about what you're working on?

Niko West 0:42
Sure. Yeah. So we're building reran, because the easiest way to look at the most concrete way to think about it, basically, it's it's two things. So it's a SDK for logging a computer vision and robotics data. And then there's a visualisation app for basically audit, taking that data that you log, automatically building like really useful, fast, beautiful visualisations that help you understand what you're building. The one metaphor that's that can be really useful is to think of it as visual printf style kind of debugging, but for computer vision data.

Jack Bridger 1:20
Yeah, that's amazing. And could you tell us the story of like how you came to build rerun?

Niko West 1:25
Yeah, sure. So actually, my two co founders and I were previously working. This was, I think, almost 10 years ago now? No, not quite a while ago, we're working at a 3d 3d scanning company that does 3d scanning of feet in a physical retail at recommend shoes. And so that's actually a surprisingly difficult problem to do really well. It's like quite quite a quite complex, a lot of things go wrong. And yeah, physical retails is a tough environment to ship like working configuration products. And so at that company, we built a really good internal visualisation tools. And I guess basically what happened to the MLR CTO, now, he came in from the gaming world and kind of approach things differently and built this like, as he started to approach problems by building this like, really rich like, visualisation, like data visualisation environments, almost. And then somehow, like, it became very easy to see what was wrong and what you needed to build from, from that, that point. And that kind of transformed how we worked at that company, I think so, so you kind of took those way, that way of working and built like some really great internal tools starting out as debug tools for for just for us algorithm developers. So that's what I was the computer vision and machine learning engineer on the team. And it's super useful for us but quickly became also than the tool that we used to kind of get the bug live devices to sort of products out. Like in production, it became used by more and more parts of the organisation. So like data labelers like Operations and Support even get used in like sales calls, and stuff like that. So just permeated the organisation was a really key part of success for that company became what very clearly one that market, let's say, like, I only got out with a good working product like yours before anyone else. And then I left and I spent spent time building mainly like mobile computer vision products of different kinds. And I needed the same thing again, it's like I had seen the light, like this is how you do things and there's nothing out there. And I kept like every new product I kept asking like okay, now someone has to have built it because it's obvious that you need it. But they hadn't. So previous to this, I started another company they like an AR mobile AR company not to get into that but the lack of great tools was like a really big problem. So that's that's just kind of like how we fell into this I decided that okay, I'm not I'm not starting another I'm not doing any more computer vision until this like tooling part is fixed because everything else has kind of become easy enough like it's relatively easy to train a neural net now and deploy it and all that kind of stuff that had been hard but just like just the problem of like understanding what your your systems are doing live kind of over time in like 3d and that kind of stuff. Like maybe on a device but also getting the exact same initialization on your development machine and the same thing on your like if you're digging into what happened on our like a big evaluation job. So that whole system was kind of really really missing. And we decided to go solid together for for hopefully to help kind of the whole computer vision and machine learning industry just are not the industry I guess but everybody doing that kind of stuff like really make good products for the make the world better. So that's kind of how we get into it.

Jack Bridger 5:03
Yeah. And when you kind of stitch it all together and look back, like it sounds like it was just this huge problem that you just kept facing again and again and again. Yeah. Was it like something that was really obvious that you had to go start a company to solve it? Or was it? Did it kind of come from looking and like introspecting? And like, was it? Yeah, the process?

Niko West 5:27
I guess I can keep going back to knowing that someone needed to solve, solve it, it had to be solved. It had to be a project, I guess we knew that. Our thought of that for a very long time, was obviously like actually solving it right, as a huge undertaking. So I think that was what sort of held me back earlier. And it kept being like, is this market big enough? Or like, is it too early and so on? And then? Yeah, I think this we started at a little bit more than a year ago. And we spent a lot of time just trying to figure out like, is this now the right time and is talking to a lot of a lot of potential users and kind of just looking at a lot of potential companies found that, that sort of all the other aspects are kind of made it mitad ripe. But now, now, people are actually building real products based on configuration and sort of similar tech, and truly deploying it and like making real things. And that's sort of starting to happen now. It's like a large scale. So it's, I think, give some more of a timing question all the time.

Jack Bridger 6:32
Yeah, yeah, that makes sense. And so when you when you decided to kind of work on it, and like, now you're building something, could you talk about, like, how you thought about what you should build in terms of like, I know, You've done a lot of thinking about, like, open source first is not open source? And,

Niko West 6:50
yeah, so yeah, sure. Absolutely. So we just had Inklings that it should be open source. And that it was, it's really important for this kind of thing. That it's something that is adopted by the developer. One, I mean, there are many aspects to this, but let the first thing that we kind of knew from experience is that for these products, like the new knowledge about how the product works is always generated by the developer working on on the product. They're trying to solve something for themselves, they understand or some aspect of an algorithm is performing in some situation, or building something new and so on. And when they do that, they make if they can, they make really useful, like explanations because they're trying to solve a real problem. They really understand that deeply. So that's where the real new knowledge actually is generated and kind of encoded into visualisations. So, it's in tools becoming a lot less useful when they're kind of made after the fact by a tools team. So that was like one, one thing that came from a lot of experience, just seeing that over and over again, that it's really important to have the developer be the one dude, building a lot of virtualizations, at least, are the researcher. So that was one aspect. Another one was really us finding that computer vision is so broadly useful, it just can, has the potential to have huge, huge impact into so many different markets. So like construction, security, agriculture, just all forms of autonomy, augmented reality, just really, really widely. And we spend a lot of time talking to a lot of these cancer developers in the beginning. And I really found like, it was, we had their own experiences from our friends, and so on, but just really seeing that people also change jobs, like easily between those industries. So lots of people change, like I'm, I work in medical AI, and next year, they're an AR, or a self driving or something like that. So this is really the same developer across all these industries. So you have this like market dynamic, where there's one layer that's horizontal, which is this developer, and then you have lots of like, application areas that are very different than if you were to try and sell kind of directly into each vertical. That would be impossible really to do. Like top down here. Like you just come with a, you know, a application focused website, and you come in with your salespeople and so on. You can just target then you have to target just the top few. But if you go above, thumbs up so you kind of the developers are adopting it. You can actually target. Yeah, all all of them. So that then it gives us another aspect that that pushed us to like, this needs to be developer focused and have bottoms up adoption kind of product.

Jack Bridger 9:50
Yeah, yeah, it makes sense. So those developers are the ones that are just like, totally, you know, they don't really need They're like super, super tailored products as long as they can, like tweak it and stuff, and they don't need tail audit, or maybe put another way they don't need tailored, like landing pages and sales people have got loads of examples of like, you know, here are the, here's, here's the success that we brought this other industry.

Niko West 10:21
Yeah, exactly. They don't need, they're not exactly excited by the white paper saying, like, Here are five reasons. And this is how you increase efficiency by 50%, or whatever. It's, they can see the capab they know what capabilities they want. And if they are, if they haven't thought of it, and they just see like a video in our example, they go, Yeah, that's fine. I need that. And that you can kind of be much more capabilities focused, which is a lot broader in our, in our case, at least, that's the really addresses a lot more people. Yeah, so that that has huge implications to how you I guess, glue can go to market with a product like this.

Jack Bridger 11:00
Yeah. So it's just been all about like getting that kind of, as you said, bottoms up developer kind of love that they really,

Niko West 11:09
yeah, it's, that's, I guess, are the first big, big part in our go to market, you could save in how we go about that we've worked really closely actually with several teams. So our initial approach was really that we built. So we wanted the open source, but when we had like, had all the the whole, everything was open source licence. I mean, they can just say something quickly about that. So the the base setup is that we are open core company. So the base base reran is completely free and open source is Licenced under MIT and Apache two. And it's just on its own, it should be like an incredible product, but it's targeted at like a single developer and be working locally on their own machine with like, or locally with one robot or one device, or that's the use case that we're really, really optimising for. And then the commercial product will build on top of that, like targeting needs of teams, working together with like products in production. So initially, though, we didn't start out targeting like single developers, it's actually harder. Just targeting a single developer, you have really high requirements on like, super low friction, it just needs to be more built out. And you need to never get stumped, like the first like, if you hit the friction point where the thing I wanted to do wasn't possible, directly jump to, I guess you'd say like a less powerful tool that at least you can kind of hack around. And if you're addressing a team, they are not necessarily like that, then you can build a subset of features that like solves their biggest needs for the team. So that's what we started with actually started working with a few teams then, and made sure that we had like, more or less, I guess, the team, your version of the product working, working for them really well. And then sort of started breaking out to make it more and more sort of focused on on single developers. And we're still really in that process of just improving continually improving the experience for single developers, sort of in that prototyping stage in particular.

Jack Bridger 13:25
Yeah, so the, the singular use is completely open source, and then the teams you, you started out with the teams, but in the long run, that part would not be.

Niko West 13:37
Well, all the things we built will be our free and open source, the team's focus things that we will charge for, or more rather than, like direct, like live collaboration, syncing data, there's a lot of issues around performance that makes collaboration really difficult. This is just this datasets get quite a big because basically what we're doing since we're not just helping you visualise maybe your dataset is really the internal state of your, of your algorithms over time. So you can have like exploding off the data lot. You can imagine, like maybe if you're doing optimization, or trying to optimise the position of camera relative to an image or something like that. Then you actually want to be able to look at the rich view of all that data for each optimizations that maybe. So that's an example that will start that initialization requires maybe 100 times more data than just input an output. Yeah, potentially. It gets a lot. And it's really important to be able to look that deeply. So just being able to kind of collaborate around that you need good infrastructure for, for sharing, and just performing handling performances is really critical. So those are some of the like more team aspects of those things we didn't build for for anyone at that time. But you can still use our product as a team, right? But just you don't get all those those great features until we can have the release the commercial product.

Jack Bridger 15:06
Yeah, yeah. Okay, that's really cool. So there's like a very clear way where you've got, like, there's very clear problems that fit into there kind of like,

Niko West 15:16
Yeah, I think so. And it's also quite some simple in other ways, just like if you need if some problem needs a service running, then that's first. That's commercial.

Jack Bridger 15:26
Yeah, that makes sense, because it's going to cost you. And then yeah, pretty

Niko West 15:30
straightforward. Like we need to run it. It's also easy to understand for users like, yeah, they're managing the service, like, course, I'm gonna pay for that.

Jack Bridger 15:39
Yeah, I felt like that's actually like a really, really important part of like, the pricing. It seems like developers are like, what, this doesn't cost you anything. Why am I paying for this? Yeah. And if it like, if it does, then they're a lot more like, amenable to

Niko West 15:53
pay? Yeah. Yeah, exactly. And I guess we as being developers, ourselves, have experienced that it's not exactly rash, or maybe it's rational, but I don't know. It's not maybe the most sympathetic, but we're like that ourselves. So we were not better ourselves. So we definitely empathise with that feeling with other developers?

Jack Bridger 16:16
Yeah, yeah, I'd like I, I totally. Yeah, I think there's times I've been very frustrated with software. I'm like, what it's literally like one tiny little thing I want to change. And now you're forcing me to upgrade? And it's, yeah. It's really, it's a really nice way to, I know that, since you kind of released that things have become like, you've been getting a lot of a lot of love online and a lot of kind of GitHub stars and activity. Could you talk a bit about how you've generated that?

Niko West 16:52
Sure. Well, I mean, first of all, I think the most important thing is that, I hope at least that we're building something that will be like really, genuinely useful, and that a lot of people are that have been in this space, I've been waiting for something like this for a long time and maybe built things themselves. Maybe they built something great. And then kind of the key sort of took its hold on that product is now dead, or they change companies, and they can't use it. So they just like, I think a lot of people are in this space. Like, I've been wanting this for a while. So I think that's I mean, obviously the the core thing, but then obviously, we did some more tactical kind of things. So our product is very visual. So we did spend, like even one, when everything was closed, we spent some time just trying to practice I guess, visually telling, explaining what what we run is, by make like making different kinds of like, videos of examples we were building, and trying to like figure out kind of how that works, what people write what resonated with people, like, just basically, I mean, none of us were good visual storytellers before, I guess, I don't know that we are now but I think we're a little better. And just going through a lot of just posting a lot of videos, and so on the kind of learning just how to explain it and what what kind of got people excited. I think that we did that continually for the first four or five months before the release, and I think that had a really big effect. Then we took all our learnings. So we open sourced, or it was always open source, I guess, but we made it public in mid February. So that's like almost a couple of months ago. Yeah. And for that launch, I guess we took all those learnings. And then I basically went like we created all the the base material and I went and sat with a real kind of editor and we tried to put together like a really tight, less than one minute video, like really showcasing and explaining how everything works. And we spent a lot of effort to make that easy to understand, but also kind of engaging and so on. And I think that had a huge effect. So that video made our kind of launch I guess go quite far. So that was like one of the things that we did and just also tried to kind of build on the open as people do. We had some some success early on. Can I just write into lots of computer vision developers like directly DMing them on LinkedIn. So that is I guess one of those like surprising things, but I guess we tried to do it in a non salesy creepy way, just sort of, hey, we're working on this. Love your feedback or some variation of that or like, I think what you're you're doing is cool, but I don't know if it's even hard to say without making it sound like a creepy LinkedIn thing, but we try to list and to be relatively genuine, and so on. And I think that was a good way to get in early conversations with a lot of like a potential users Yeah, no, relatively okay. It's not scalable, but it's the easiest. It's fairly easy, at least. And it was useful. And I think that that was also good too. Like we got, I guess, in front of a lot of people. People like that as well. Yeah. Yeah. early on.

Jack Bridger 20:17
Yeah, that's really smart. And I know you kind of you said that, like, a lot of the people that are on LinkedIn, and do use, it may not actually be, you know, on Twitter.

Niko West 20:30
Yeah, yeah, I think, I guess the tech community is like, very Twitter focused. I mean, I, I also follow things on Twitter and so on. I'm not I'm not great at it myself. But there are a lot of great people in tech or that are not on Twitter, really, I don't know, maybe follow, like, just, they have an account, but it's not important to them. So in particular, I think that this like little bit more hard sciency fields, so I can robotics and computer vision can have. Their more I found that many of them are more likely to be on on LinkedIn. And they find lots of good papers and like interesting methods, like there are hundreds of the influencer sharing stuff like that that are I mean, they're really doing useful job sharing. Interesting work. So they're on there. And it's a little bit overlooked by a lot of the tech community because it's, I guess, not as cool kind of yell arbitrage.

Jack Bridger 21:24
Yeah, that's really cool. I guess it's like, kind of this platform is not inherently bad. Like people. There's Yeah, if there's good people on there, and you're, you know, and they do go there for some reason, then, yeah,

Niko West 21:39
absolutely. Exactly. I guess the third thing was, I mean, we are a rust shop. So we all have, we run this building really from the ground up in rust. So that's like, we use Apache arrow. But other than that, like our whole database is custom, like from the ground up, like the renderer is written on top of W GPU, but not it's a completely custom, we use the rust GUI framework YUI. So it's like, really end to end rust. And that's important for us for lots of reasons. But rust is also a good it's just a really popular area of rust developers love rust, and love talking to us with other people. So that has been an important driver for us to just get out in front of lots of lots of people. It's just sort of talking about that we're doing it in rust and why and so on that that's been kind of a big driver for for getting exposure for the project for us as well. An interest in so on.

Jack Bridger 22:40
Yeah, yeah. And it also helps with your hiring.

Niko West 22:44
Yeah, yeah, I just found this. This continually. Every time we've been out hiring. I just I'm amazed by the incredible talent that is ready to switch jobs for just just for the sake of working in rust and their day job. Just so many great engineers and I it's kind of in one way gets makes me sad that like I can't hire all these like great people. Maybe I mean, we're a small company. We don't can't do the old school Google thing and just hiring everybody. That's good. We need to kind of match a particular role and all that stuff. But yeah, all teams out there that are that could be using using rust and are having trouble hiring and they're really missing out. I'd say that's been super useful for us.

Jack Bridger 23:28
Yeah, so So building rust.

Niko West 23:32
I think you should build and rust. Yeah, if at all practical. It's a very strong, strong choice. Put it that way.

Jack Bridger 23:41
Very, very cool. Yeah. And Nico. I think that's we've got time for but I wondered what would be like your key takeaways for any founders of dev tools, or people working in Dev Tools listening?

Niko West 23:54
Sure, I'd say the main thing is that does work for us consistently and continues to seem to work well is that we're building we're solving our own problems. Are they sorry, I mean, I've experienced these problems for so long. That I just have internalised what is needed. And I think several other other people on the team are in the same situation. So that's just helpful all the time. Sometimes when you're building open source in the beginning, it can be harder to get get like rich feedback, because it's open source. They don't need to ask you for anything, they can just use the product but we always have that guide of we know exactly what to build because we're we're solving it for ourselves. And worst case, we fail on that now we have like amazing, amazing tools to build something else with. So that I think it's just incredibly useful every day. And that that tip is hard to to like overdue, I think.

Jack Bridger 24:51
Yeah, that's amazing. I absolutely love it that you're just on this mission to like never do another computer vision project until we have this amazing tool. Yeah,

Niko West 25:00
it just never again.

Jack Bridger 25:04
Yeah, let's say great. And where can people learn more about rebrand and about Nikko?

Niko West 25:11
Sure. I think the easiest is just our website. So that's rerun.io. You can find links to our Twitter and different things like that. I'm not the most the I'm not the biggest Twitter account. So maybe just go there. Or find us on Twitter. I think it's our handles reran.io just in regular language. So one of those two, I think is the best.

Jack Bridger 25:37
Basic. Thanks so much for joining Miko, and thanks, everyone for listening. Thanks