We’re with the team f'{ing_python}’, that finished third in the HackAUBG tournament. Their idea is simple – to make your life easier. “Parkinc.” is an app that can show you vacant parking spaces, so you won’t need to circle around with your car, searching for a free spot. Stefan Andreev is the organizer and he also works on the Machine Learning part of the project with David Atanasovski. Stoyan Lisichkov is accountable for the web app and the whole front-end. Eva Konevski is working on the design and also the front-end. Last, but not least we have Dimo Rezashki and Christian Deyanov, who are developing the economic and business model of the project and how to make the idea profitable.

Can you tell us more about Parkinc.?

Christian: When we were brainstorming ideas and how to improve the city life, because the topic of the hackathon was Smart Cities, we came up with this idea, because it’s really close to us. As we know, in Sofia there is a big problem with parking spaces, especially in the downtown area, because they are reserved for the ones who live or work there. So our idea was to make the process of finding a parking space much easier and faster, so people don’t lose their time, energy and money and pretty much – make lives easier.

How does it work from a more technological point of view?

Stefan: Basically me and David focused on the ML model, using PyTorch. We feed an image, taken from a stationary angle from a CCTV camera, then we would pick out spots from legal parking spaces and feed this to a binary classification model, which would evaluate whether or not the space is occupied or vacant. We designed this to connect and update the counters for vacancies through a backend SQL server, which we had connected to a web socket to our website that Stoyan had made.

How do you develop such a thing in just two days? Was it hard?

Stefan: Basically David and I did a huge Crash Course on the PyTorch library and it was a very rudimentary model since this was my first time making a successful convolutional neural network. But once you have the basics done, you just have to play around with the parameters and get your model to work at an accuracy that you desire. For our case it was around 85%.

David: Also, because it’s a binary classification model it’s really easy to make it very close to a 100% accuracy, because it has only two outcomes and if we had more data, pictures of cars and vacant parking spaces, we could have done this model to work at 99% accuracy.

And how do you test the whole thing?

Stefan: In ML you have the three splits – the test set, the validation set, and the training set. The validation and training set work with each other while the testing set validates the results, using loss functions for training and validation. It’s all in the code and for us it is kind of intuitive.

Were there any challenges during the hackathon?

Stefan: Finding the data set and normalizing it for our standards. Originally, we found a data set, but it was not in the proper resolution to feed into our model and I had to resize all the images, because build-in library functions were not working for resizing. Those were the issues with the model.

David: There was also an inconvenience that some of the libraries that we used for the model like the Python libraries and everything else that was required for the environment, worked together with certain versions, so if there was a newer version of a particular library that is used in the software, it did not want to work with the other parts of the project, because it was not implemented for that yet. So we had to find the correct versions, the data set and normalize it. For us they had to be 150×150 and labeled too. So that was the challenge. Our model is 85% accurate, because we could only find 12 000 images of free and busy parking spots. There was actually a way way bigger data set – about 150 000 images, but the images weren’t labeled and we couldn’t really use it for a ML model. And now here are the challenges from the web developers.

Stoyan: The only challenge that we had was how to send the user data from our web app to PyTorch and get back the output to the people so they know how many spaces are there. Initially we did a web app, because in two days it’s very hard to do a cross-platform, so to do an iOS and Android in that time is very hard. But we managed to do it all, and everything worked, the socket worked, so basically it’s a done project. We just need the cameras.

Eva, can you tell us about the design part of the project?

Eva: We worked on it together with Stoyan and I also worked on designing a logo. I wanted to implement a new logo for our app and for that I used Illustrator.

Stoyan: For the design I used mostly CSS and JavaScript to make it look really nice.

Eva: We also used jQuery for some parts.

Stoyan: If you’ve seen our design you know that it’s simplistic, but it brings a nice aesthetic feel.

What benefits do you see in your product?

Christian: Our whole business and economic model was built on the fact that our service would be free for the mass of customers, which would be citizens of the big european cities. We know that american cities were built for cars and car usage, so they don’t have the infrastructure problem with the parking spaces. For example european cities like Sofia, really struggle with this which is a real problem for normal commuters, who probably go an hour early to their workplaces, just to make sure they have a parking space free to them. Our whole idea would be to add value for the commuters by saving their precious time. It also saves stress, because they don’t have to worry if there’s a vacant spot for them and also as an added benefit, we believe that our app can drastically shorten the commuter time spent in cars around the city center areas, that will also improve the air quality at least to a certain degree.

Will you continue to work on Parkinc. as a group? And what future do you see in the product?

Christian: Our product for now is very unique, because I think there is only one project like this, which is completed, but only in the Netherlands. So there is really a big gap for such products to start and actually provide value to the community. To continue our project we need funding, because the software is there and working – front-end, back-end and etc. Everything is nicely set, so maybe we only need bigger data sets and the infrastructure of CCTV cameras. When we were considering our business model, we did research on CCTV cameras and how expensive they are, and we concluded that with good funding we can make our project work in the foreseeable future. So the only thing for now that is stopping us from continuing improving and working is basically funding.

Were there other projects on the hackathon that you liked?

Christian: Personally, it was the project GeoShare that won the hackathon. The other one that I liked was Udoma – they won the second prize. I think that both of them can in a way communicate with our project and make a city really smart and efficient.

What do you think about the startup ecosystem in Bulgaria?

Christian: I think that here we have the environment to actually do startups. There are a lot of unexploited markets and fields of improvement. We definitely have the human capital of smart IT guys like my colleagues, who are able to build sophisticated ML Algorithms and code in a short period of time. We have the technical skills, we have the space to actually try something new. But the only problem in general would be the funding, because as far as i know there is no state funding for startups and any big commercial fund that would actually invest in startups in Bulgaria. But – yeah. Everything is here, except maybe money and publicity.

Final words from you guys? A conclusion?

Stefan: Basically I gathered these people for their talents. Two business model developers, two web developers and two backends, including myself. We all contributed and we ended up developing our idea to the specifications and to the scope of what we intended. If we had developed too niche of a problem, it wouldn’t have made sense economically.

Ain’t that the most important thing, that you’re not just developers? You all have a set of different talents.

Stefan: Indeed. We have to take into account all the aspects of the project and not just the development. Because David and I were part of the same team who got third place last semester. But we only did the development part and bombed the entire economic and design portion, because it was our first time attending.