The 80th episode of Datacast is my conversation with Alberto Rizzoli - the co-founder of V7 Labs, a platform for deep learning teams to manage training data workflows and create image recognition AI.
Our wide-ranging conversation touches on his entrepreneurship background since university; his experience participating in Singularity University; his work developing the Aipoly product to aid the blind and visually impaired; his current journey with V7 Labs to enable any business, large and small, to leverage the sense of sight to automate any visual task; lessons learned from fundraising, finding customers, hiring; adoption of AI at European enterprises; and much more.
Please enjoy my conversation with Alberto!
Listen to the show on Spotify, Apple Podcasts, Google Podcasts, RadioPublic, iHeartRadio, and TuneIn
Key Takeaways
Here are the highlights from my conversation with Alberto:
On Building The Entrepreneurial Muscle
Entrepreneurship has always been something I have wanted to do. Both my grandfather and father were entrepreneurs, in the paper publishing and cinema/TV industries, respectively. They moved forward in the world of media, which was not something that always fascinated me personally. But I have always wanted to put my own stamp on something and drive my own path forward.
During university, my first entrepreneurial endeavor was to work in a 2-person team on a 3D-printing startup. 3D printing was hot in the early 2010s, which showed many promises. Two things that I learned from that experience are (1) tackling a market that will enthusiastically buy your product right away and (2) working with bits, not atoms since physical products take a lot more time to be built than software products.
In entrepreneurship, the pros outweigh the cons in every way. As a startup founder, no matter how good or bad your startup is, you learn three times as much as working at a company — especially when you are young and aren’t given big responsibilities. If you have to build everything yourself, if you have to sink or swim, you will learn things that you would otherwise not because they have to be outside of your comfort zone. You have to go and learn how to fix something if it breaks because everything depends on it. These are the things that you can happily do in your 20s, make mistakes, and live off very cheap budgets.
On the other side, working in an enterprise, especially a good one, can give you a lot of learning on how things are done at scale. Sometimes, I find myself hiring people who know a lot more than me about how something should be done in a larger company, which is great. But at the same time, I wish I had a bit of that exposure. I think the best of both worlds is to always have the learning attitude and talk to many people who work at other companies that may be doing something (which you are interested in) better than you.
I was also fortunate because I could start a business economically. Even though I lived on very little money, I had the means of doing so. I had my university pay for my scholarship. I was also privileged growing up. If you are able to take home these adventures and do them, then, by all means, do so and always do it seriously to win in the end.
On Singularity University
Singularity University was a project founded by Peter Diamandis and Ray Kurzweil to make a new type of university. The 2012–2015 years might have been the best time for doing something like that. All the technologies that today everyone talks about were just emerging. We had CRISP discovery and self-driving cars in 2015. We had deep learning with the first AlexNet paper in 2012. There was this incredible hope for a new decade of technology.
It was an incredible experience, probably the three months that I remember most fondly in my youth. We were put in an isolated NASA base with 70+ ambitious people. Great stuff happened from there. I developed friendships that will last for my whole life, even though it was a relatively short timeframe. When you are stuck in a foxhole with people, you really bond. Secondarily, it was one of the best times to look at technologies and imagine their future. Some turned out to be false promises; some turned out to be extraordinary (like neural networks are continuing to revolutionize the software industry). I hope that there will be more programs like it to create more communities, especially as young people increasingly have to forge their own path and think about a world that is increasingly uncertain with changes every decade.
For context, we would have guest lectures from Silicon Valley people who have started Google’s self-driving car project or pioneered neural networks. We would have the chance to debate the future of that piece of technology with them. That’s pretty invaluable, especially for the confidence you receive — meeting face to face with the people who have started your fields and understanding that they are just a bit older and more experienced, but you can still sit at the same table if you have the ambitious, curiosity, and politeness to do so.
On Aipoly
I was genuinely excited by the possibility of bringing AI to inaccessibility technology. The inspiration came from a childhood friend of my parents who lost his eyesight in a hunting accident. When you are permanently blind, you find things a lot more challenging than just being partially sighted or visually impaired. There’s not just an element of navigation where a dog will lead you to where you have to go. There’re also the elements of narration and understanding of what’s going on around you. We wanted to train a classifier on so many objects that can be a descriptor of your life, which led to Aipoly.
There is a dichotomy of entrepreneurship that the world doesn’t openly discuss. Sometimes the most loved companies are the least successful ones. Aipoly was a very loved company with a loved mission, mainly because it helped one of the most vulnerable populations out there. Aipoly was an award magnet. We would tell our stories and warm people’s hearts everywhere we participated. But sometimes, the best stories also don’t have that strong of a market behind them. The challenge on the other side is that there is less of a support network (effectively capitalism) in helping these companies grow. How can companies with more of a social mission find self-organized support like private companies do?
On The Inception of V7 Labs
In 10 years, every software company will need a training dataset to achieve its goals and will run a lot more on learned approaches than on code. The purpose of V7 is to make people not worry about training data anymore. Suppose they want an AI service to perform a task to detect something. In that case, it should be elementary for them to add raw examples and label them without worrying about the enormous costs and time sink of building a training dataset. V7 is an automation-focused SaaS product that is above and beyond the best user experience in the market. And I say it with a bit of bias because I made it. But it’s something we have put a lot of love and attention into. The public reception so far has been staggeringly good. We are in the fun hyper-growth phase of a startup, in which the customers pull it harder than you can keep up with certain things.
We started in 2018. It’s been a pretty amazing journey because our customers are primarily AI scientists, who are some of the smartest people that we have met. It’s always nice when talking with people working on new ways of solving something with AI and applying software to previously unsolvable problems.
The meaning of V7 came from our visual cortex. The human brain has six of them, each with a simple shape. We discovered that the visual cortex works in layers, and similarly, we build neural networks with layers. We wanted to create something that was the next level of the visual cortex — something that could enable machines to learn as well as humans, maybe even better than humans at some point, and create this new Cambrian explosion of AI companies.
On Finding A Good Co-founder
Co-founders are so important. There will come a time in which either you are sick, you are bummed out, or you are tired of something. They will be a mirror to you and provide the help/support that you need in order to go through the hard challenges. Most importantly, they can complement your skill sets that breed a unit. Eventually, founders are kind of like a one-person thing, especially in the early days of a startup. Both should have the technical skills, but one may have deeper technical skills, and the other is more of a storyteller. That’s the dynamic that Simon and I have.
We met at a hackathon. We literally sat at the same table and built a crappy game engine to play video games on the public TVs at pubs. We made it in 24 hours without sleeping and had so much fun that we continued to tinker with stuff. Simon may be practically a part of my family. It’s fortunate to find someone you also never argue with since that’s usually an indication of a good co-founder when you can maintain a professional relationship with and discuss even the hardest stuff. If we ever have an opportunity to start something again when V7 IPO and we are older/more successful, then I would definitely turn to Simon again for an adventure. But we’ll see if we even have the energy for that in the future.
On V7’s Product Capabilities
Annotating data is the unsexy part of machine learning but also the one that eventually makes your model work well. Even if you work in an unsupervised problem with unsupervised data collection, there is still the element of being able to organize that in a way that works well. We wanted to create an experience that is delightful to use. With V7 Darwin, you can throw images at it and then define a few labels in a semi-automated way. You define a region of interest (like a coarse box), then it will find objects within it that you can segment in a class-agnostic way. Today, you can start labeling a dataset and train state-of-the-art computer vision models on that same dataset within a few minutes. This helps teams that label data on their own. The effect tends to be not lower training data cost but rather an enormous amount of training data. Other than that, the annotation experience works like a high-end graphic designer tool, which is influenced by our backgrounds a bit. We want it to feel just like a professional design tool for labeling machine learning data.
Dataset Management is one of the most at-face-value boring features we have, but also the hardest one to create. Personally, it’s my favorite. You can see all of your datasets in one place. You can manipulate them and view the results of your annotation at any time. You can also view the history of every single annotation stage that it’s gone through, as well as the amount of work that has gone there. Our back-end engineers have made a product that can handle billions of images robustly and hardly ever slows down. Suppose you have a gigantic dataset with hundreds of labels across hundreds of classes. In that case, you will easily be able to filter them and create search queries to assess the data quality, then assign tasks for the annotators. We wanted to create that fluidity since working with open datasets comes with so many mistakes, primarily because you can’t visualize the whole data corpus easily. We wanted something in which you could see all the data, unpack the ugly bits, and weed them out.
With V7 Neurons, you can train instance segmentation models with very little training data. My favorite client use case comes from a UK-based company that assesses the health of poultry so that we can have free-range poultry the world can afford and no more caged hens which suffer all their lives. It’s an AI to understand whether the chickens are living healthy, eating/drinking enough, stretching, etc. This was all done using models built by V7. They initially labeled 60+ images. The resulting model ingested those images and was able to understand all the activities that the chickens took by rendering 500 polygons in real-time. Not everyone uses our Model Automation capability because some teams are incredible at building their own, but our models are nonetheless helpful for creating more training data.
On Fundraising
You, the founders, are going to be the differentiator. You need the funds to make it happen, so raise early and fast. Seeking investors is a lot like dating: go for the ones where the resistance isn’t particularly high. There’s a lot of people with money out there who are willing to invest in talented people. The one thing you need to do is build trust. Your number one priority is to look for people who think and act a lot like you in some ways, then pitch them first. The complement of your board will come with time. If you are in machine learning, you are in luck because it’s obviously a hot space for investment. Don’t try to branch too far from your geography. It will be a lot easier to find people in the city or country you are in before going to Sand Hill Road.
In the UK, a lot of VCs have backgrounds in finance. While in the US, there are more foreign entrepreneurs and outlandish idea funding. I found that there are more requests for proof-of-value in metrics and money in the UK, which can be hard to find. Especially in ML, a lot of the grunt work goes into your product development's starting point. It might take months before the thing actually takes off because it needs to learn enough. I’m still in London, the best place in Europe to seek funding.
On Hiring and Culture Setting
One great engineer or data scientist counts as much as 50 average ones. So go for the smartest people that you know. Also, be generous with equity and salaries. Finally, if you have a goal that is crazy big enough, people will automatically be attracted to what you are doing (especially if you talk about how to help them build up their careers by pursuing your goal).
Your idea is probably not as ambitious as it will be when you get started. When you get started, you try to find a product-market fit. From there, things begin to evolve almost magically. You have a longer-term vision of where things will go and the world you want to build. This is the story that you should tell as the founder. You must be the one who looks miles ahead and imprint the excitement into your potential hires.
We are lucky to see AI applied in use cases all over the world, from the plantation of fruits in South America to the extraction of oil & gas in Australia. So our need is to have a team that can relate to global people. On the other side, they need to be able to relate to nerds because our customers are data scientists who like talking to someone who’s a bit like them. These are our two cultural fitness: are you able to be beloved by our community and by the varied representation of our community?
If culture is the sum of your past experiences and backgrounds, we obviously have many things to represent over time. I don’t think the ML community is restrictive when it comes to backgrounds, considering emerging talents all over the world. Open education in the field really makes a change for a bigger difference.
On Finding The “Ideal Customer”
You should have a combination of problem and customer when getting started. You want to make sure that the person having this problem is someone you really like and want to talk to every day, even when this person is really annoyed at you, to get a good customer fit. Simon and I spent a lot of time brainstorming and taking days off to think about who we wanted to dedicate the primer years of our life towards. The ML community is a good one because they are a lot like us — facing the training data problem. Anything that swerves you away from serving that customer is a mistake, and you should say no to opportunities like that. No matter what the upside is, no matter how tough the time is now, if you stay true to your one mission, this simple playbook tends to work out.
For us, we need to understand what problems an ML scientist in specific domains has. In the industry, a scientist is given a pile of garbage data and told to shovel it into an elegant model. This person would tell us a ton of problems that they have. We spend a lot of time talking to customers in various industries with different problems, then put them all into a gigantic sheet of issues and find a single system that could solve them all.
On The MLOps Ecosystem
I think the MLOps space is quite nascent and will come from win-win collaborations. The great ML stack will consist of startups focused on their domains making the best possible X, Y, Z for ML (such as model pruning, model versioning, model inference, etc.). It’s a bit like other forms of SaaS, though we are yet to see the full evolution of MLOps.
We should start collaborating on standards; otherwise, we end up in a standards crisis. Annotation standards are an absolute mess. I hope we get together and figure out an adjacent standard that works for everybody.
On European AI
I think the enterprise adoption of AI is lower than in the US, but so is everything in software. But I see projects that are sometimes more impressive in scale/magnitude than in California. There’s a bit of tradeoff (similar to the VCs), in which Europeans are much more structured. When the project starts, it starts seriously with no R&D pilot phase. We are also more cautious in jumping into new technology, so people still use traditional ML in production.
There are also other things in Europe that have to do with our legal requirements in handling data, but I don’t see that as a differentiator because it’s just manipulative. There’s a scientific hunger in European founders because we have a tendency to work on humble projects that might not be flashy but extremely difficult from a scientific point of view.
On Public Recognition
Recognitions are a red herring. If you are out there and see your friends getting them, consider that awards are just as much of a business as yours, and don’t focus too much on them. They’ll come when you do something flashy and cool, but sometimes you’ll be the most successful you’ve ever been, and no one is going to notice — and that’s fine.
In general, I like that many awards we received were for the positive use of AI, which is a sign that the world wants to see this technology go in the right direction. The awards that I’m most proud of are the ones that I haven’t won yet and hope to achieve with the work that we are doing. So I’ll keep my head down, keep our team happy, and have us marching forward. It’s a very, very long road for all of us ahead.
Timestamps
(02:14) Alberto briefly shared his upbringing and education at the Bayes Business School in London.
(04:01) Alberto shared key learnings from his first entrepreneurial stint at 19 by developing a 3D printing product for ed-tech.
(07:48) Alberto described his overall experience participating in Singularity University’s Graduate Studies Program at the NASA Ames Research Park under a Google-funded scholarship in 2015.
(12:52) Alberto helped develop the Aipoly product to aid the blind and visually impaired.
(17:38) Alberto showed his enthusiasm for federated learning applications within mobile devices.
(19:53) Alberto talked about the dichotomy between capitalism and social good in entrepreneurship.
(22:29) Alberto shared the backstory behind the founding of V7 Labs.
(26:40) Alberto discussed the comparison between biological and artificial neural networks.
(28:02) Alberto emphasized the importance of having a good co-founder.
(30:27) Alberto dissected the notable features developed within V7’s Annotation capability.
(33:37) Alberto went over things to look for in a video labeling tool, citing his blog post.
(37:21) Alberto unpacked key principles behind V7’s robust Dataset Management tool.
(40:53) Alberto walked through the powerful capabilities of V7 Neurons that power its Model Automation tool.
(43:33) Alberto shared fundraising advice for founders seeking the right investors for their startups.
(46:07) Alberto shared valuable hiring and culture-setting lessons learned at V7.
(50:12) Alberto emphasized the importance of not losing sight of the ‘ideal customer’ for young founders in the AI space.
(53:01) Alberto shared the hurdles his team has to go through while finding new customers in new industries.
(55:10) Alberto walked through labeling challenges dealing with medical imaging datasets.
(57:35) Alberto discussed outreach initiatives that helped drive V7’s organic growth.
(59:49) Alberto mentioned the importance of collaboration between companies within the MLOps ecosystem.
(01:02:01) Alberto touched on the scientific hunger of Europe regarding the adoption of AI technologies.
(01:03:49) Alberto briefly mentioned what public recognition means to him in the pursuit of democratizing AI for the world.
(01:06:07) Closing segment.
Alberto’s Contact Info
V7’s Resources
Mentioned Content
Articles
“7 Things We Looked for in a Video Labeling Tool” (Aug 2020)
“The Biggest Mistake I’ve Ever Made: Losing Sight of the Ideal Customer” (March 2021)
Talks
“An AI Narrator for the Blind” (TEDx Geneva 2016)
“If The Blind Could See” (TEDx Melbourne 2018)
People
Geoff Hinton (for rethinking the ML field fundamentally)
Chelsea Finn (for her work on meta-learning)
Jeff Clune (for making agents that work at scale in the real world)
Book
“Start With Why” (by Simon Sinek)
Notes
V7 is hiring across all departments. Take a look at their careers page for the openings!
About the show
Datacast features long-form, in-depth conversations with practitioners and researchers in the data community to walk through their professional journeys and unpack the lessons learned along the way. I invite guests coming from a wide range of career paths — from scientists and analysts to founders and investors — to analyze the case for using data in the real world and extract their mental models (“the WHY and the HOW”) behind their pursuits. Hopefully, these conversations can serve as valuable tools for early-stage data professionals as they navigate their own careers in the exciting data universe.
Datacast is produced and edited by James Le. Get in touch with feedback or guest suggestions by emailing khanhle.1013@gmail.com.
Subscribe by searching for Datacast wherever you get podcasts or click one of the links below:
If you’re new, see the podcast homepage for the most recent episodes to listen to, or browse the full guest list.
