…as a data scientist, you need to have a mindset of a software engineer first… There are a lot of tools right now that are on the market… But the last thing you want to do is reinvent the wheel and build something that someone started two or three years ago…
An interview with Paul Pop, co-founder and the CEO at Neurolabs, a computer vision platform for automation tasks that raised €1.2M from 7percent Ventures, Lunar Ventures, and other investors.
Peter Zhegin:
Good morning, good afternoon, good evening, depending on a timezone you're in. So we are all very welcome to this talk. My name is Peter Zhegin, I'm talking to Paul, who is the CEO and the co-founder of Neurolabs. Neurolabs is the company that focuses on synthetically generated data to help develop object recognition models. Hi, Paul, it's a pleasure to have you here with us today.
Paul Pop:
Hi Peter, first of all, thanks for the invitation. And thanks for the lovely intro that was very much spot on. So I think you know, we can jump right into it. I'm one of the three co founders of Neurolabs, where we do synthetic generated data for computer vision, specifically for object detection. That's the one thing that I will add to what you said.
Peter Zhegin:
It will help if you give us some background on yourself, and maybe a bit describe what Neurolabs does actually, and how does it help data scientists and data engineers.
Paul Pop:
I'm a computer vision engineer, I spent roughly the last 10 years of my life working on algorithms used to recognise objects in images. And through this period, I came across one recurring problem, over and over again, and that is the lack of quality training data.

At Neurolabs we created a no-code platform, that's aimed at making computer vision available for a non specialists. So for RPA developers, software engineers with no computer vision background. And we provide the trained computer vision algorithms that we train in virtual environments. We put together a simple interface where as a user, you select your 3D models from a public catalogue or from your own private catalogue. And as an output, you get a data set of generated images in 3D in virtual environments. But on top of that, you also get a state of the art networks that are trained on that dataset to the best of our ability.

We also give you the weights and some wrap up code in order to allow you as a user to embed it into your end application.
Finding an idea for a startup – reading reports vs. talking to customers
Peter Zhegin:
I feel that a lot of data scientists sometimes think about building some tools, but definitely not all people make this step towards building a company out of that. So were there any particular moment when you realised that you were ready to launch a startup?
Paul Pop:
Yeah, it's a very, it's a very good question. So we're not a classic example of, 'I had a brilliant idea that can solve an end user problem'.

My co-founders and I, we came across each other in the university and then after about 10 years, we came to a point in our lives where we were kind of fed up with our day to day jobs, one of us finished their second Master's because you can never be overeducated, as Oscar Wilde puts it right. And the other one had enough of his job in a bank. So we decided to start something in AI. We didn't really have the idea from the beginning. And we slowly went through the path of doing some small projects left and right up until we came up with the idea of synthetically generate the data for computer vision specifically.

We all have a little bit of tangents with this topic and our background. For my experience, particularly, I used to work as a computer vision engineer at Huddle in London. And we used to do player tracking for sports, specifically for football players. And my colleagues were playing FIFA and it really made sense - why don't you just use FIFA as training data? And why do you have people labelling when you could just get that synthetic data out of there for free?

There was no one particular moment when we said, 'this is this is it, let's start doing'. It was very much a search and exploration for us.
Peter Zhegin:
You highlighted a very important point of ideation. So you said that there was kind of a process, right, when you looked at multiple ideas. Sometimes people play with different things, pet projects, but how they can progress through these things, and how they can select something that is actually might become a potential startup?. What was your heuristic, a framework to do that?
Paul Pop:
I have many ideas here. Some of them, looking back were wrong. But I'll share with you a couple of them. So at first we looked at McKinsey reports as to where do we need to apply AI as a community in the next couple of years. And we came up with the retail sector as one of the most underserved but with highest payoff in the long run. So we started looking at use cases in the retail as to how can we implement AI in there. And then we got to know a lot about the retail sector, but not necessarily about a good problem that fits to our skills. So I wouldn't necessarily go down that route again.

One thing that I learned during this journey is how to ask questions that give me insights and not questions that give me answers I'm looking for. One great book for this is 'The mom test', I highly recommend that book. And for us, it was a little bit of a game changer, because we saw when we're doing some user interviews with professionals in the retail sector, we were influencing them into saying one thing or another that we actually wanted to hear. But we didn't actually came to the bottom of their problem. So I really recommend that book.

Another thing that I would say comes as a natural step in this direction is going through an accelerator. So once you have an idea outlined, and once you have at least an MVP, you should look to go through an accelerator. We went through a Fast Track Malmo, and it was it was great. People over there really put us on track and really gave us this steam to keep the engine running. I would definitely recommend that route to jebra, who's looking to start something new.

Peter Zhegin:
And when you were moving through this ideation, this filtering process, at that moment, did you guys already quit your jobs?
Paul Pop:
We took that quote from Arnold Schwarzenegger - 'Plan B? There's no plan B'. So we did quit our jobs, we said, 'let's make the best of it'. It aligned interests a lot more. Because if you have some people with a job and some people being completely out of it, you are soon going to start to see some friction just because someone is at a much higher speed than the other person. So, at least two of us were unemployed at the time. The other one was finishing his PhD, so almost there. But we did go with no safety net.
Collecting customer feedback and sticking to your vision
Peter Zhegin:
Could you guide us a bit through the sequence of building stuff and then commercialising it?
Paul Pop:
I'm personally a strong believer in a thin slicing, start with the minimum viable product and be diligent about what that means. Have a devil's advocate to look into every little thing that you could put in your MVP, and ask, does this really have to be in there is this really an MVP? Don't try to be greedy from the beginning - do a thin slice, put it in front of users get their feedback and build something they want, not something you think that they want. So that's one. One thing that I strongly suggest, get something in front of users. as well for feedback, make sure that you're not influencing them, and you're actually getting what they really think about your product, not what you want them to think about your product. Work with that, start building on that and keep going towards a product.
Peter Zhegin:
An interesting moment here is how to filter out some feedback that maybe is not relevant, and how to find the balance between you as a founder, and what potential customers want. What's your approach to that?
Paul Pop:
We put together a spectrum of end users. On one side, specifically, in our case, you have computer vision engineers. And on the other side you have the enterprises. The more you go towards the enterprises, the more the customer will have one specific need, that would feel like they're pushing you towards a consultancy project that is good for them, and only for them.

And the more you go towards a software engineer, the less the need for customization is, the more they can do themselves. And all they need is one little bit of software that works really well for one specific use case. And they'll build the rest. So striking the balance between the two, it's still something we're working on. But we see that going from enterprise to solution providers, meaning companies who build software for enterprises, and we come in as another piece of software in their offering. And then we go another level towards computer vision engineers and write about in the middle, we look at RPA engineers as one user persona for us. So these are developers that are pretty skilled at putting code blocks together, but they don't necessarily programme the code blocks themselves. So Neurolabs can come in, and it's one of these code blocks as a part of the workflow for what I would call a business developer an RPA developer, in one because most of these people have some understanding of the business.

And in between the RPA engineers and computer vision engineers, you have software engineers who are very capable of writing software, they don't have the computer vision expertise, they take a Neurolabs network, they embed it into their work.

On the end of the spectrum, we do have computer vision engineers who are very skilled in computer vision. They write the latest state of the art architectures, and they change the networks themselves. So what can we do for these guys? Well, for these guys, we're providing datasets. For these guys, we're providing an extra level of data, that it's pristine data, it's pixel perfect, annotated. So it's a, it's a lot of data that they can use on top of their existing datasets in order to make a great product.

This is our spectrum. This is how we define it. It's different for everyone. In my experience, the more you go towards a problem that you solve, the more you have to build custom software. If you go a little bit more towards the other end of the spectrum, you have to do one thing really well and you have to be one of the best people are doing that. So it's a tradeoff between the two where your company lies, I guess it's up to you to decide.
Peter Zhegin:
What's your approach to defining things that you, as a founder, or as a founding team, are not moving from? And you just stick to them?
Paul Pop:
Yeah, that's, that's a really good question. And I wish I had three bullet points answers for you. The truth is that we all gave up our comfortable lives in our little cushion in London to do something that we like. And that is to to implement an AI solution in the real world. We don't want to get diverged from that too much. So we don't want to go too much into building something custom for the enterprise. When we had conversation with investors who said: 'Oh, but you actually have some traction in the retail sector, you should double down on that and just grow a company out today'. And it didn't feel right for us. It could be great as for a company, but just not for us as three founders who started out looking a little bit more at a computer vision engineer end of the spectrum. So that that's one of the things.

We also are quite strong believers in remote culture, not necessarily fully remote. But today, we have two offices, one in Edinburgh, one in Cluj in Romania. And we are a believer that this is something that can work for the future, you do have one office that you belong to you, you go there and maybe two days a week or something like that. Now with the pandemic, it might be even less, right. But you have the freedom to operate on your own. And you have enough guidance, if needed, by going into an office and talking to people face to face and having that water cooler chatter that sparks ideas when necessary. We're believers in getting smart people next to you, smarter than you. And if you put them there and let them operate autonomously, they will provide you with great inputs and with great work. You just need to set the table for them. So I guess these two are a couple of the things that guided our journey so far, and we look for investors who understand them and are behind them.
Peter Zhegin:
I think you mentioned very important point, the team, the culture. Can you add some background on your c- founders?
Paul Pop:
The three of us are coming from Edinburgh University. That's where we met. We all done computer science, mathematics and physics and the mix of the three. My two co-founders have worked in banks, JP Morgan and Bank of America in London, technical roles, some maths and statistics and some software engineering. As a founding team, we're very technical. And one might say we lacked on the business side. That was the challenge for us, from the beginning to have that business mindset and stepping away from your keyboard and your dev environment and figuring out if you're actually solving a problem, or you're coding because you have this one idea that should work, but it's not validated by the real world. For us, it was it was a bit of a challenge in the beginning. I think it's still is as we grow, but we're managing it better and better with the type of talent that we get as an addition to the team, and the type of people that we get to complement our skill set.
How technical co-founders can deal with sales and other commercial tasks
Peter Zhegin:
And how did you overcome this challenge at the very early stage when there were just three of you and you needed to get this first talks with customers, first interviews, what was your approach?
Paul Pop:
I'll tell you what we've done. And I don't like it before going into the things that we've done, right. We've done quite a lot of things that worked suboptimal. You had the three of us doing a little bit of the tech part, the three of us doing a little bit of the business, a little bit of networking, a little bit of strategy. And it's normal in the beginning, but then you should specialise quite fast, you should look at being very good on one particular thing. And even if your expertise is not really aligned to that one of the cofounders needs to be the business person, you just need to assume it. There's no way around it, I don't think there's a silver bullet that can solve it. One of the cofounders needs to assume it, needs to look at how to monetize the whole thing, at what customers saying, how do you get their feedback, how do you keep them happy, and how do you build lasting relationship with them.

It took us quite a while. That's the reason I'm rolling a little bit on the subject, it took us quite a while until we separated the rows clearly enough. On paper, we did it probably from day one. But in practice, it didn't really happen that way. We took a long time before we decided that - Okay, now you consciously put down the MacBook and stop coding and you go talk with with a customer or you go to that conference, or you have that presentation.

Peter Zhegin:
So the idea is that you need to specialise within your team, even if your backgrounds initially are more or less similar you still need to specialise. And you need to train and increase your expertise in the field where you lakc it. In your case, business stuff commercial things.
Paul Pop:
Yeah, I think I think that's something that we should have done better. And that's something that most companies that have technical background should think of in the beginning. For sure you benefit from hiring someone early with complementary skills. However, if they are not a co-founder, I don't think they can make up for it. I strongly believe that one of the co-founder needs that responsibility from the get go.
Peter Zhegin:
Yeah, that's very interesting point. And from what I see, as an investor and what they hear from other founders and investors, it's critical to have someone among the co-founders on the commercial side, because it allows a better insight into product into customers heads.
Paul Pop:
That's an interesting thing. Yeah. Do you think that you could align the interests of say the third or fourth employee, with the co-founders, I mean, of course, you would give away some equity of course, it would get further Way, a good financial package? Yeah. But do you think that you could align the interest and the motivation in some one way or another
Finding first investors
Peter Zhegin:
And since we talked earlier about investors, how did it work out for you guys? How did you find your first investors?
Paul Pop:
The very first investor was an angel investor. Then we had the accelerator that invested in us and alongside came another angel investor. And roughly a year after we started going after VCs, we picked up two angel investors on the on the way, we try to optimise for smart money. So people who are advisors as well as an investors, or they open up some doors for us into strategic places. This year [2020] in February, March, we were supposed to wrap up our round, and then the lockdown came, it was a little bit of a setback for us. But by that time, we already had the lead investor Lunar Ventures from Berlin. It was a little bit of a setback, as I said, because the other investors in the round were a little bit hesitant at the time. And we postponed the round closing up until June, which turned out to be early July. And we ended up having two other VCs coming along 7% Ventures from London, Techstart from Edinburgh and Northern Ireland. And also id4 Ventures, a fund between London and Paris, and alongside two other angel investors.

For us, it was good to take a breather, let the dust settle a little bit and then come up with a stronger round. But we did go through some rough patches in March and April, when was a lot of uncertainty. We knew we had a lead investor, but we didn't know exactly what we wanted to close anymore. Was it 12 months runway 18 months 24. It was very uncertain at the time. And also, what type of burn do you want to have? You want to hire more people? Do you want to stick with the same team? A lot of questions that had to be answered and the answer was just give it some time to settle and then to come up with a better strategy. I think we're in a great position right now. And we managed to sail the waters happily. It did cause a lot of grey hair for a few of us. But you know it is what it is at the end of the day.
Peter Zhegin:
I'm curious, these first angel investors, where did they come from? Is it your personal network, your ex-colleagues, maybe through university?
Paul Pop:
That's a good question. For us, it was not our immediate network, but the extended network. We didn't know these people from the get go. Actually one of them we did, but the rest of them came from people who knew that we were starting something and said, 'Oh, I have this one friend that might be interested'. We got in touch, and we clicked. Also through the accelerator, we got to meet a lot of people, a lot of VCs and a lot of angel investors. I will make an emphasis on that - go to an accelerator if you can, because it really changes the way your company will develop in the future. And I think VCs all of them have one or two favorite, angel investors alongside them, who are pulling tickets that are significant.
Peter Zhegin:
And just being cautious about your time - just two other questions before we close. We talked about many different things, but maybe I did miss something important. So maybe you want to talk about something else that I didn't ask?
Paul Pop:
One problem that I we faced as a company was fundraising on a technical product, that is not a one liner, it's not, you know, 'Shopify, for schools' or something that you can just make it catchy and explainable in one line. So we had quite a technical product that targets a particular technology, computer vision, in this case. The challenge was, how do you explain this to investors? How do you put it in a pitch nicely? It didn't help that we all had technical backgrounds, and none of us worked in television before. And we struggled mightily with this part.
Peter Zhegin:
How did you solve it?
Paul Pop:
We got some pieces of advice in the beginning that raising funds from VC, it's a numbers game. I disagree with that advice. But in practice, what we ended up doing was contacting a lot of VCs that weren't necessarily looking for investments in our area. But I do believe that, when targeting VCs who are looking at deep tech investments, you don't need to target 100 of them go for maybe 10-20. But know them by heart, know them what they invest in, what they believe in, why would you be a fit for investment. And also keep in mind that they're going to be with you for 5-7-10 years. So the last thing you want to do is, to throw away some proposals and just pick randomly one of the VCs who actually wants to invest in you. So be mindful of the people that you're targeting. Sometime, the pattern of 12 slides is not the best for you, when you come to the problem slide it takes a little more than a sentence to explain it. So find the people who understand that and don't waste your time with those that don't really look into that.
Closing remarks – a mindset of a software engineer, a business person, and a data scientis
Peter Zhegin:
Yes, I can definitely relate to that. That's a nice comment. And the last question from me would be - a single piece of advice for a data scientist who is thinking about a startup?
Paul Pop:
I think as a data scientist, you need to have a mindset of a software engineer first, if that makes sense. There are a lot of tools right now that are on the market. And it's very, very, very hard to keep up with all of them. But the last thing you want to do is reinvent the wheel and build something that someone started two three years ago, and it's quite far down the product market fit path. So think as a software engineer and try to solve the problem, as if you didn't know anything about data science, and you were looking to pick tools off of the shelf. If you can't find that, then it might be that you're getting to the right time. Or it might be that the problem is not there yet for the larger market.

It's very tricky to think as a data scientist, and as a business person at the same time, I think you need to put another hat on and try to validate that what you're solving is a problem for a big number of people. Is it 1000? Is it 10,000? A million? It's hard to gauge but try to get the order of magnitude fast. And once you've got that you multiply it with what's it worth for every person that you're solving that problem for? And that gives you an answer if you should spend the next five, seven years of your life building that product or not, I guess.
Peter Zhegin:
That's very interesting and useful advice, I'd say. Thank you, Paul, once again. I am sure we will hear a lot of good news from Neurolabs, and I'm sure we will hear a lot of positive feedback from data engineers and data scientists who are using or will use your tools.
Paul Pop:
Thank you for the opportunity. Peter. One thing I would like to add, we do have a private alpha coming up. So if we have some listeners who want to have a go at object recognition using synthetic data, give us a shout Neurolabs.eu, and we'll get you on the private alpha. Will really appreciate some feedback and some word of advice as to where to take our product next.