We talk to startups and investors, you get the value.
Rocket DAO ecosystem
Visit any place where you can buy wallpapers and you will see thoughtful people walking among the shelves. They are solving a difficult intellectual task: they are trying to imagine how their flat will look like judging from the one bolt of wallpaper on the show-case. With the help of Wizart technology, you can instantly see the result on the screen of your smartphone by pointing the camera on the wall and trying on the bolt of wallpaper you like. Vasiliy Yavorchuk, the CEO of the startup, tells us how they managed to create a solution for which customers come on their own.
We talk to startups and investors, you get the value.
I have been working in the IT-sphere for more than 20 years now. In 1999, I graduated from Grodno National University, got a degree as a mathematician-economist and started working in a bank as a programmer. Then I became a partner of a beginning outsource-company, I also was a project manager and run two companies: Intexsoft and Exposit - an outsource development, we completed more than 200 projects for clients from 60 countries.
There were different clients among them - from game-dev to implementing ERP-systems in large enterprises. I have so much experience in technology and so many managing practices that I’ve started feeling a bit bored. So this is how I came to the thought that I have to create something interesting and technologically innovative.
I finished a hands-on course for company presidents in Skolkovo to find new values and tasks for myself. And at the same time, one of our clients - a huge wallpaper distributor from Canada, made a complaint that there were solutions for visualizing paint on the photos but still there was no similar soft for trying wallpapers in a real interior.
My colleagues from Exposit and I found this idea quite interesting. We did some deep research and found out that there was a hypothetical solution for this task - neural networks. In a few months, we created a prototype and went to a large industrial exposition in Moscow, where we showed it to a dozen companies.
It was a bullseye: some of the wallpapers’ manufacturers, especially from German, have been unsuccessfully trying to find a solution for visualizing wallpapers in the finished interior. We saw a demand that no one could satisfy.
For sure, we had to develop in this direction, so I started forming a team for Wizart development. The team includes my colleagues from Exposit, we found a few people with amazing expertise in machine learning, product management and sales. Currently, there are 22 people in the Wizart team. At one time, I had to use my charisma to the maximum to convince them to participate in the project at the early stage with vague prospects. Perhaps now, nobody regrets as we are doing a super-important thing and it seems that it is in a great need.
It is enough to take a photo of your interior and choose a finish-view of your interest from the catalogue in the application. The system will automatically change the finish-view and it will be crystal clear whether it suits your interior.
We strived to simplify the process for the end-user. You don’t need to do any space measures or configure additional settings. Just download the photo and get the result.
To get such a result, it is necessary to create a 3D-model of the accommodation from a 2D photo. It is done by our neural-network converter, and each of them does their part of work.
Long story short, this is the algorithm: at first, we define the geometry of the accommodation, where are the corners, walls, floor and ceiling. Then we define where the user is in this geometry, and at which angle he looks at walls. It is also important to find out how the light falls, its direction and location of the sources, so we can produce a realistic picture.
At this stage, we get the model of the accommodation with white balance, user’s location, location of the light sources and other information. After this, we rebuild the model with a new texture on the walls and floor, which is then shown to a user in the application. Calculations are made on the cloud-server or directly on the mobile device. We have applications for iOS and Android, and a web-client, too.
We started developing Wizart three years ago. Back then, computer vision technologies were much less developed than today. There were simply no suchlike solutions; so we made them ourselves from scratch. There were no platforms, no datasets, no specialists - nothing.
That’s why we had to conduct many pieces of research on our way, so we could understand what can be realised from the technical point of view. We turned to the work of researchers from technical universities and came up with new ways to apply their best practices to our context. In the process, we created new practices in machine learning; no one has done this before.
It took more than two years to create our authentic computer vision technology. It consists of a complex set of neural networks of various types and algorithms, and each of them solves a separate mathematical problem of converting a two-dimensional photo and transforming information from device sensors into a three-dimensional scene, geometric dimensions, shadows, colour balance and so on.
The absentation of the ready-made solution in this field has become our main technical challenge. It is interesting to create a technology of such level but it is also difficult and time-consuming. These conditions save us from competitors - it won’t be cheap and fast to create a product similar to Wizart.
The finishing materials market was something completely new to our team as we never worked with it before. We didn’t have a clear idea of how to pack the product and prove its value. We were urged to spend a lot of time to understand the market rules and establish a liaison with major market players. We participated in industrial expositions and conferences, communicated a lot with stakeholders and tried to sell a product that didn’t exist.
Another challenge for us was the fact that there are no ready-made datasets. We had to create ours from scratch, remodelling it several times and realizing our own mistakes in the method of selection, layout, balancing. We tried to work with outsourcers who prepare customized datasets by the hands of people, mainly from Asia and Africa. We were not satisfied with the quality/ so we returned to the markup and did it our way. In the process, I came to a very important insight - it is better to have a compact but well-marked dataset, than a large one and of a low-quality.
It is possible to get a decent result with a relatively small volume of data if you have good expertise in machine learning. But it is easier to process specific exceptions and problematic cases on a small dataset - you need less data, containing this exception, to affect the behaviour of the neuronetwork.
We position Wizart as a B2B solution. It integrates into websites of manufacturers and sellers of the finishing materials helping to save customers from painful choices and thus increasing sales. Monetization - a subscription with payment for the period.
We started to sell our product in two markets: the German market as I know it for quite a long time and there are a number of top-list manufacturers in the country, and the Russian market which is mentally close to us. Last year we had the first implementations in these markets and it wasn’t easy.
We had started working with our clients two years before we realised the product, we held dozens of meetings, had been trying to resolve contradictions for a long time and traded to death. It is always hard to work with first clients.
This winter we took part in a major trade fair in Frankfurt. All the largest wallpaper manufacturers exhibit their novelties and we went there with our modest booth. We are interested in companies from other regions, not only from Germany. That’s how we appeared in the market of the USA, Turkey, Central America and the Middle East.
After the first implementations, customers come to us on their own, because they see our solution on the site of their partners or competitors. So one exhibition gave us a pipeline for the year ahead.
Such interest is also heated by the fact that only one company in Canada makes a similar product, but we have better visualization quality and it is more convenient to implement our solution. Soon, we will add a “secret sauce”, which will raise the value of our product significantly. Still, we do not disclose it, but we’ll present it soon.
Now we have 12 paying customers, the plan is to multiply this number in a year. Revenues will grow by multiple times as we have large companies in the pipeline. There is one idea how to increase our lead stream tenfold, but as for now, we won’t give any details, this is our know-how.
We’re raising a $400 000 seed investment round to scale the sales and implement new product features. Our perfect investor is the one who has worked in the construction and maintenance field or solved similar tasks. Especially entering large retail with his own IT-solution. Expertise in computer vision is also “smart money” for us.
We have more than 100 ideas in the backlog, one of the breakthrough-type is using a light radar which Apple is actively implementing to their products. With this radar, it is possible to create a more detailed accommodation model and give better results in the end. Also, we’ll automatise a common user-case - the selection of wall and floor finishes from the photo. To achieve this, we’ll add integration with Pinterest so we could offer the user interior options based on the pictures that he liked. You can immediately transfer your favourite combination of materials to your interior and quickly find the best option.