Wed, Dec 4, 2019
It all started with three friends and a crazy idea, and ended up in a machine learning consultancy with more than 50 contributors.
This year, that crazy idea turns 10 years old and behind it lies an exciting, fun, nerve-wracking and inspiring journey.
There's no better time to take you on a walk down memory lane and dig up some juicy stuff about the origins of Tryolabs and its evolution.
We'll give you a behind-the-scenes look at our story, including our favorite highlights, takeouts from the entrepreneurial days and, who knows... maybe some previously unpublished top-secret information. Also, we'll cover a timeline of high impact moments in the field of artificial intelligence (AI) and how they affected our work.
Let's get started! 🚀
Our story begins with Martin, Raúl and myself, Ernesto. Raúl and I are working at the same company and discover a great dynamic working together. We then meet Martin and immediately feel that something great could happen if the three of us start working together.
We decide to quit our jobs and do our own thing. Inspired by the few Silicon Valley AI startups around, we start working in stealth mode on a product that requires all of our skills as a team. The what of the project is not as important as the fact that we're doing it together.
Our office is the barbecue station at Martin's place. Counting with (almost) no roof, the working space is hot in summer and freezing in winter, which requires us to move close together and celebrate team spirit.
ImageNet is quietly created. It's a large visual database envisioned by Fei-Fei Li from Stanford University, who realized that the best machine learning algorithms wouldn't work well if the data didn't reflect the real world. For many, ImageNet is the catalyst for the AI boom of the 21st century and for us, it will have a big impact in our decision to start Luminoth, our open source computer vision toolkit.
We keep invading unorthodox office space, this time in my grand-parents shed (which some people say was haunted, and it most definitively was). It's a cozy 100 square foot office with three desks, cobwebs, some heavy metal and a lot of love.
The vibes of these first months as a company: a lot of risk and a lot of promise. The motto is Enjoy the ride! as we care much more about experimenting and having an awesome experience, than achieving specific results.
It's during this time that we start working on Ansguru, a product built in our favorite language (Python, of course). The app uses AI for classifying people expertise areas, using their questions and answers. It was just the moment that Ansguru could revolutionize the world of tech. Spoiler alert: it didn't.
Kaggle, a website that serves as a platform for machine learning competitions, is launched. Our team starts looking at those challenges to keep learning about the field.
Enjoy the ride is not enough, at least not if we want the company to survive. Apart from the product we're prototyping, we manage to get some clients for Python development projects. This means that we're now working on some cool initiatives for startups, which keep the machine going and help us grow as a company as well as professionals.
Luckily, we get accepted at the Ingenio business incubator, where we finally get a decent office and a group of mentors to torture with our problems.
Being part of a community of startups helps us keep going in the hard times and allows us to share every moment with people that are fighting the same fight. Also, it helps us hire the first people, which turns out to be the best decision ever. The team starts creating the culture and work dynamic, which will still be part of Tryolabs almost 10 years later.
Not only the team is expanding, but also the tech community we're engaged with as a company. We co-organize and sponsor events, such as the first Python Day Uruguay or DjangoCon US 2011.
Also, it's the first time we're able to afford a company wide dinner party (for 8 people and not too fancy).
Using a combination of machine learning, natural language processing and information retrieval techniques, IBM's Watson beats two human champions in a Jeopardy.
Ansguru, our first product, dies, but we keep exploring product opportunities.
Why? Not because we have one-in-a-life-time product ideas, but because we feel that it's not about the product. We know there's a great value to be added to the industry using the technologies we love and we know we're the right team to do it.
Our team keeps getting bigger and with joined forces we launch DayWatch in LATAM, the first real-time daily deal analytics tool. It's another experiment on our own IP and focuses on the power of AI for the e-commerce industry.
At this time, after a year of advisory meetings and fulfilling discussions, Eduardo Mangarelli, Director of Technology, Developers & Platform Group from Microsoft Latin America, joins our board as Advisor and Investor.
PD: If you're curious about our crazy website and the many inconsistencies in our offering back then, you can check out our first website here and learn what not to do.
The computer scientist Alex Krizhevsky designs the convolutional neural network AlexNet. Using GPUs during training, the model significantly outperforms traditional image recognition models in terms of efficiency and speed. The research reshapes the computer vision landscape and is motivation for the convolutional neural networks we later on use for building deep learning object classification, detection and segmentation models.
Pivotal year. We keep seeing a lot of product opportunity in the natural language processing (NLP) space and decide to kill DayWatch. We start MonkeyLearn, which (spoiler alert!) in the following years becomes a totally independent and very successful company.
As a consolidated company, we now manage to keep growing and target bigger clients. We find our true value proposition that helps us position Tryolabs as a team of machine learning experts. We start solidifying our knowledge and expertise through exciting projects, which paves the way to where we'll stand in 2019.
Few months later, we win the award for the Most Innovative Exporting Company in Uruguay, which marked one of the heaviest celebration nights to date! Some people are still trying to recover to this day.
A team of researchers led by Tomas Mikolov at Google publishes and patents the paper Efficient Estimation of Word Representations in Vector Space (Word2vec), which proposes novel models that are used to produce word embeddings.
This is going to be proven really useful in some of our NLP related projects a few years later.
Up to now, we were focusing on NLP projects, but due to some clear industry hints, we decide to start exploring computer vision as a potential offering. This includes the re-evaluation of our product/market fit and training ourselves in new exciting technologies. It's the beginning of our love story with computer vision, which will five years later be at the heart of our services.
Similarly, the trainings we establish in this year lay the foundation for our culture of continuous learning and experimenting with new cool technologies. While it's still early for deep learning in production, it's one of the main fields that we're prototyping and doing research in.
In the meantime, our product, MonkeyLearn, is doing well and we isolate a part of the team to work on it full time: our spin off company is officially born!
The former Google Brain researcher Ian Goodfellow and his colleagues invent the generative adversarial networks (GAN). Given a training set, this technique learns to generate new data with the same statistics as the training set. This facilitates the generation of pictures that look very authentic when compared to the original ones. At Tryolabs, we start doing research in this field and building prototypes with GANs, which aim to solve real-world business problems.
While MonkeyLearn is accepted in 500 Startups batch 13 and becomes an independent company, Tryolabs keeps growing.
Our network becomes stronger and most of our clients now come from referrals. The word of mouth effect is priceless and helps us get Tryolabs more and more recognized in the AI space and in the competitive Silicon Valley market.
After a lot of heated discussions about how cool it would be to build some very strange prototypes and applications, we decide to host our first internal hackathon. A space for collaboration and team-building with the great excuse of building some innovative and no-strings-attached projects.
The hackathon offers the great opportunity to challenge our creativity and play with technology in a way that we just can't in a high stakes scenario. Read here the full story about our first hackathon.
However, the hackathon isn't enough when it comes to innovation and we aim higher. We start working in stealth mode on a computer vision toolkit. The goal is to understand computer vision problems end-to-end and to build an open source tool that solves the most challenging parts when building a computer vision model, so developers can focus on getting their code into production without the model building hassle. This is the first step towards Luminoth.
Google's AlphaGo program becomes the first Computer Go program to beat a professional human player using a novel form of reinforcement learning. At Tryolabs we benefit from the same advancements in reinforcement learning and build our own pricing agent, able to determine the best price for a product.
We launch Luminoth and our international appearance starts to boom. Thanks to the open source toolkit we're invited to talk at conferences around the globe, such as the ODSC in London and SF, and reach a new record of frequent flyer miles among the team.
What started as an R&D project and a tool to be more efficient in our work is now an effective marketing tool. Luminoth is actively used and liked among the community and generates international visibility for Tryolabs.
While 2017 is a usual busy year with amazing projects and clients, we manage to work on some cool initiatives for the team (second hackathon and a working trip to NYC) and for the local community (sponsorship of our first scholarship for university students).
A great recognition for our effort comes from the Uruguayan Chamber of Technology, who awards us as the most innovative company of the year.
GANs, invented in 2014, make great progress and are now used for image enhancement focusing on realistic textures rather than pixel-accuracy. This leads to higher image quality at high magnification and the first faces are generated. A few month later, we'll be using GANs to build deep learning models able to automatically retouch images.
Sebastián Gonzalez, with whom we've been having a long lasting professional relationship, joins our team as the new CEO. As it would be for any organization, this is obviously a major shift.
Sebastián brings a fresh view of the company, which now counts with over 40 people. He helps us think deeper in how to solve our problems and prepare for scaling.
We professionalize our processes, improve our structures and prepare the company for upcoming challenges, always having in mind how to better serve our clients and give more and better opportunities for our team.
We start investing much more in internal trainings, focusing on technologies that are key to the projects we tackle. We curate training material for new members joining our team, such as the AWS workshop, which we also open source.
In the meantime, we keep crossing the continents to visit our clients and spread the good word of AI. Some highlights include:
Researchers at Google AI Language publish the paper BERT (Bidirectional Encoder Representations from Transformers), which presents state-of-the-art results in a wide variety of NLP tasks. Leveraging the innovative technique will help us win the #AllianzGIHack the following year in Hong Kong.
An amazing year with one highlight after the other. While most of clients and network were so far based in the US, we start increasing our presence overseas.
Our CTO and two machine learning engineers from our team travel to Hong Kong and win the 48-hour #AllianzGIHack. Shortly after, we start working on a machine learning project for our first client based in Asia.
Another great honor comes a few month later when the Uruguayan Engineering Association recognizes Tryolabs co-founder Martín Alcalá Rubí for his contribution to the business development of AI software from Uruguay to the world.
2019 is also a remarkable year for AI in Latin America. Khipu, the biggest AI gathering in LATAM, takes place in our hometown, Montevideo. Excellent speakers, such as Jeff Dean, Yoshua Bengio, or Nando de Freitas share their experience at the conference and as proud supporters of it, Tryolabs is represented with two talks and a booth. For Khipu attendees and friends, we organize the unofficial Khipu after party, which turns out as an unforgettable event with more than 300 attendees.
Khipu proves how the AI community in LATAM is growing and gives us great hope for the future of AI and its collaborators.
With more data collected than ever, network capacity and processing power are pushed to the limit, and new solutions for on-site data processing are required. It's the rise of edge devices and big players start to launch hardware accelerators, such as the NVIDIA Jetson Nano or Google Coral Dev Board. We start exploring the power of these devices and research their suitability for specific processing tasks. We present our results in an edge device report, which you can access here.
The last 10 years have been quite a ride on a rollercoaster and looking back, we can certainly say that it was worth all the ups and downs. Counting with more than 50 collaborators, top-notch partners and a big network, we're proud of where we stand today and of having found our spot in the machine learning world.
All this has only been possible thanks to a very passionate team of talented engineers who contribute every day with their experience and expertise. Thank you Tryolabs Team, you're the best!
Also, we're very grateful for everybody who stuck with us during the last ten years and helped us become the company that we're today. We feel fortunate to have fantastic partners, collaborators and friends that facilitate our work and inspire us. Thanks to you guys, we could learn a ton and develop ourselves and our expertise over time. Thank you! ❤️
Counting with such an amazing network of people, exciting machine learning projects in the pipeline and great events coming up, 2020 promises to be another great year for Tryolabs. We're by no means planning to settle down and will keep challenging ourselves and aiming higher with new technologies and projects.
Cheers to another 10 years! 🍻