The history and goals of BroutonLab
Hi, my name is Mikhail Yurushkin, founder of BroutonLab. While ago it struck me an interesting thing. Many public persons talk about the advantages of neural networks. Everything looks easy. What you need to do is to clone the pre-trained neural network from GitHub and finetune it on your data. But almost nobody talks about arising problems. As a result, the hype around AI obscures people. They spend money on the development and training of the neural network to follow the trend. If they do not have some kind of AI in their project then their projects become not interesting. As a result, founders often lose their or investor's money, time and interest in this area.
To somehow solve the lack of information problem, I want to write a series of articles. I will share practical tips about the development and integration of AI in real life. I hope they will be useful for entrepreneurs and founders of technology startups. Founders of many companies that I had to work with face the same difficulties in the field of AI. And I will cover them in my articles.
As part of this post, I want to talk about how I founded the BroutonLab and its goals.
Before the foundation of the company.
During my studies at the university, I worked as a C++ software developer in the company. In parallel, I worked on my Ph.D. dissertation. The fields of my scientific research were optimizing compilers development and high-performance computing. I spent several months on the development of a fast matrix multiplication algorithm. Finally, it was faster than Intel MKL on Intel's processor!
But at some point, I needed for an area of expertise changing. After 7 years of work in the field of software development, I was completely burned out. There were no vacancies related to my research work in my city (even now!). I did not consider moving to another city or country due to family reasons. It was interesting for me to work on problems that involve my background in mathematics. To my mind, it’s strange to learn mathematical disciplines for 6+ years and then throw this knowledge. The only way out for me was freelancing and searching for interesting projects.
Around the same time, I began to take a closer look at other areas of IT. And my choice fell on machine learning (deep learning has not been so popular yet). It was both interesting and required mathematical background. By that time, I had already attended a machine learning course at my university and I knew what it was. Even so, I have done one of the most important stages of my education. I entered YSDA (Yandex School of Data Analysis). This school is very popular and prestigious in Russia. In the year of my admission, 3000 people took part in the admission process. And only 120 of them went to school. I was one of the luckies to pass the entrance tests and interviews. Studying at this school formed the basis of my expertise in the field of data science. I was ready for more difficult challenges.
The first orders and the BroutonLab foundation
I began to receive my first serious orders in the field of data analysis while working on my Ph.D. dissertation. I will forever remember the work in Opening - the Irish startup. I used Deep Learning and NLP here for the processing of resumes and vacancies. It was a very interesting experience, which only reinforced my desire to work in this area. After the moment I defended a dissertation in applied mathematics, I had more time to work. In 2017 I registered as an individual entrepreneur. Approximately this time I started to think about the data science company foundation.
I worked on the Upwork exchange and looked for orders that were interesting to me. In a certain period of time, the number of my customers became more than 2. It was difficult to help them. That's why I started to organize a team of data scientists around me. The first members of my team were Anya, Juri, and Lev.
To sum up I will provide the dry figures: we started working at the end of 2017. In 2018, we rented an office of 60 sq.m and there were 6 of us. At the end of 2019, we moved to an office of 100 sq.m. And now our team is 14 people.
During this time, we have completed a large number of projects. And we faced a lot of difficulties. One of the main problems was that many of the projects did not overlap with each other. For example, there is little in common in
- automatic analysis of scientific articles using NLP methods
- processing of satellite photos using OpenCV
- handwritten text image processing (ICR problem)
I have been working in data science projects for more than 3 years already. I must admit that the development of artificial intelligence is an underspecified problem. There is still no clear definition of the term "deep learning". Every year, yet another new state of the art solution comes out and replaces the previous one. Even working full-time in this area, it is impossible to keep track of all research papers. All this research becomes more and more like a snowball.
In such a situation, it is very easy to throw money away. Especially without fundamental knowledge of data science. I will provide some situations that can affect the budget very much:
- choosing the wrong architecture of a neural network
- trying to replicate arXiv paper that contains errors
- lack of knowledge in the area of feature engineering. This is a data preparation process that is important for the development of good models.
- the hiring of data scientists for full time. Such kind of specialist will cost a lot of money.
The main goal of BroutonLab is helping companies to overcome these difficulties. We are doing what we like and while working we are increasing our expertise in this area. If you love numbers, our total experience in AI development is more than 30 years in total.
The results of our work bring a state of satisfaction and pride in the work done. It is great to realize that thanks to you, the company
- received more value (for example, it automated a part of the business and saved money) or
- developed a high-quality PoC and received a new round of investments
This gives a tremendous surge of motivation for further self-improvement.
Thank you for reading this article. If you have any questions or suggestions for me, feel free to ask me them! I will answer each letter. Besides, I will be glad to receive feedback and suggestions on the content of the articles.