How to quickly find and not lose AI and Data Science specialists

In collaboration with Anna Perova

Introduction

Every day, humanity creates, uses and stores huge amounts of data. Every article, post in a blog or instagram, every Like and indeed every fact of communication - data that, when processed, becomes valuable, brings profit and warns against the risks of the one who owns them and knows how to extract the relevant information.

With the growth of data analysis and awareness of the usefulness of existing archives, the need for experts in Data Science, machine learning and artificial intelligence (AI), able to work with the data and create on their basis useful models, as well as self-processing systems and forcing them, increases. work.

Why is it necessary for those who are recruiting teams in this area to think about new recruiting methods?

As back in 2015, they wrote on TechCrunch, according to Mckinsey , who, admittedly, were not far from the truth, 490,000 specialists will be required in this area by 2018.

If you rely on LinkedIn data - out of 236 million profiles, about 11,400-19,400 are Data Scientists profiles.

Already, Amazon's average annual investment in AI Hiring is $ 227.8 million , while Google's key competitor investment in AI hiring is $ 130.1 million . Experts in the field of artificial intelligence leading companies receive from $ 100,000 to $ 500,000 per year. This is evidenced by data from the survey, which conducted The New York Times, and in principle is checked periodically comes across either on dice.com, or on monster.com, or on LinkedIn.

The area is new and in trend. The quantity and quality of young specialists does not satisfy the highest need for them in the whole world, as well as here in Russia - here the situation differs only by the order of salaries and by the number of open vacancies in the field of Data Science & AI.

According to the analysis of hh.ru, the number of open vacancies in the field of Machine Learning, Deep Learning, Data Science: more than 1000. The number of ready-made specialists with the necessary experience is not more than 300. Candidates with at least minimal experience in this area of ​​AI, Data Science are not suitable for these positions are about 3 thousand. And this in itself is a problem to search and hire because:

• on the one hand, there are really few valuable specialists;
• on the other, there are many candidates who are just starting their way in the area in question, whose training (in the case of hiring) will have to be invested.

All this leads to an extremely overheated labor market, and hiring in this area must take into account a number of factors:

• the highest competition for talents (salaries & conditions) - there are more vacancies than candidates, but the requirements for candidates are high; approximate statistics: 10-15 job offers per candidate with 3+ years experience in Data Science & AI;
• companies are forced to be more flexible in terms of salary, schedule, additional opportunities, in general, a preference is given to flexible schedule, part-time, the need for freedom to display creativity to find the best approaches and solutions;
• candidate projects and tasks are important because Data Scientist - often has a certain personality type: an analytical mind, a motivation for intellectual and professional development, a craving for research, various tasks, curiosity, and at the same time there is some individualism and demanding recognition of the results;
• a company nevertheless needs a strong team capable of delivering results on a deadline in which it has someone to study with, with whom it creates research projects;
• necessary resources and power, good hardware, GPU.

Due to the high competition for talents in this area, a number of selection questions arise, the main of which are:

• Where to find AI & Data Science specialists?
• How to recognize? How to choose the best or most promising (from a small group of candidates (who will be trained quickly and profitably))? What should be the selection criteria for a Headhunter specialist?
• How not to lose? How to keep AI & Data Science specialists?

1. Where to find?

In addition to the standard and well-known sources, I would like to draw attention to the most productive in terms of my personal experience in hiring AI & Data Science specialists.

• Slack, Open Data Science Channel. This resource is not for recruiters and is mainly intended for communication of engineers, experts in the field of Data Science.
  What you need to do: place an ad in Slack in the community of Open Data Science. It is better to ask your colleagues to do this - DS specialists or Data engineers, without hiding the level of salaries and opportunities for development. Emphasize the features of exactly attractive tasks and projects, technologies that can be used.

  
  

• Competitions Kaggle.
  What should be done: Select the top - 50-100 in Kaggle competitions. The first 20 usually solve problems for fun, work with pleasure in large companies and do not look for work. After the first 20, you can select potential candidates with high potential in DataScience and AI, contact them, suggest a meeting and a project. In case of refusal, it is possible to request recommendations using the referral program of your company (in more detail about hunting with the use of Kaggle, you can ask questions in a personal, or, if there is interest, we will prepare a separate material).

  
  
• H-Index. Hirsch index, but rather a method of assessing / searching candidates, which is better to use it when searching for AI, ML / DL, Computer Vision, Data Science experts . This criterion makes it possible to evaluate which of the scholars and professors are quoted better and who are worse and find those who specialize in the desired professional field and can become a guru for young specialists. What you need to do: look for Data Science and AI specialists using open data on the Hirsch index. Be interested in topics that fit your needs. The average index for scientists of different levels:
  
  1. young scientist, graduate student - 0-2;
  2. PhD - 3-6;
  3. Doctor of Science - 7-10;
  4. Member of the Dissertation Council - 10-15;
  5. world-famous scientist, chairman of the Dissertation Council - 16 and higher.

Useful site for searching candidates by citation index: eLIBRARY.ru.
This site contains publications of Russian scientists. There are more than 24 million articles posted, the database is constantly updated.
One of the main lafhacks is to register on the site, then find a professor with a large number of publications with a high level of citation, find a way to contact him and ask for recommendations from co-authors and students. As an option - open publications and connect with collaborators through available social networks.
When hiring scientists, it is important to bear in mind that they may lack practical skills, an understanding of business, but perhaps their scientific career may be useful for the development of high-tech projects, including in the field of AI.

• Organize your own Data Science competition: hackathon, programming olympiad. Such events make the AI ​​Community, Open Data Science, etc. You can try to organize it yourself, but the quality is likely to suffer.
  An example of a good competition: Sberbank Contest .

  
  
• Start a free ML / Deep Learning course; the format is not important. The main thing is to decide on the subject and tasks, monitoring the most appropriate specialists according to the results of the "homework" solution. For a good funnel, invite more than 50 of the most promising. As a result, about 10-15 will remain, and you will hire no more than 5, but you will save a lot of time and energy with this method.
• The system of internal recommendations. Assign a decent referral bonus to internal employees. Stimulate them for recommendations.
• Develop your AI networking. The AI ​​and Data Science community in Russia and in the world is still very small and actively communicates at conferences, it is easy to get recommendations from gurus and speakers, it is often even possible to do it for free (OpenAITalks, Skolkovo Robotics, NIPS, ICLR etc.)

2. How to select really good Data Science & AI specialists

For HR, it is not easy to understand all the concepts at once, so the most important thing is to understand the main headings well in order to at least somehow navigate. And to act in accordance with the instruction (chapter “FINAL LIST, or the Principles of Personnel Selection”) - i.e. very clearly balance the complexity of work and trials with financial and non-financial motivation.

So, for a start, it is important to determine what is now understood by Data Scientist

Data Scientists use statistical data, machine learning and analytical approaches to solving critical business problems. Their main function is to help organizations transform their volumes of big data into valuable and efficient models.

They should know mathematics well, program, develop machine learning algorithms for automation of algorithms. They are also expected to have a high ability to interpret data, it is important to be able to visualize them, problem-solving skills are important, even if the problems are not fully worded.

It is important that they can work with different types of data and data of different levels of readiness.

A good mathematical background (knowledge of linear algebra, analytic geometry, probability theory, and mathematical statistics) is a must. And this is even more important for data analysis than engineering knowledge. Learning ML models requires an understanding of which particular models need to be used, how to interpret, and how to improve the results.

Knowledge of programming languages : Python or R (but to navigate the technological stack you used); C / C ++; Java
Skills : Scala, Apache Spark, Hadoop, machine learning, deep learning, and statistics.
Optional : Tensorflow, PyTorch, Keras, Caffe, Pandas etc., Jupyter, and RStudio., Experience with highly loaded systems, Cuda.

The difference between Data Scientists and Data Engineer is the ability to not only analyze data, but also integrate them into existing systems. In this connection, deep knowledge of programming languages ​​is particularly important, as well as the experience of creating or participating in the creation of highly loaded, multi-threaded systems, etc.

Key concepts with which it is desirable to be familiar to a recruiter: Machine Learning, Deep Learning, Data Science, Data Mining, Computer Data Recognition, Car Recognition, Face Recognition, Health Systems, Natural Language Processing, kaggle contests.

Filtering candidates based on telephone HR interviews:

1. It is important to understand how deep the candidate’s knowledge is in mathematics (linear algebra, probability theory)
2. What frameworks does it use? A diverse experience is welcome.
3. What are the most complex projects you had to create? What was the personal role and result?
4. What competitions did you take part in?
5. Are there articles in scientific journals here at habr.com?

Algorithm of recruiting and selection of candidates:

1. The technical interview consists of 3 parts:
   
   • Online testing for 20 minutes. An example of a site for placing an online test. ;
   • Testing - 1 hour. Technical interview in the office. Test task 20 min-1 hour. You can create a test of 10-15 tasks (problems in probability theory, mathematical statistics, computer vision, machine learning). The test is performed by the candidate alone in the meeting room. He does not have to solve all the problems, but it is important to solve at least 50%. In testing it is useful to put points for an objective assessment and the ability to compare candidates;
   • The oral part of the technical interview is 1 hour (discussion of the results of problems in probability theory, mathematical statistics, and analysis of how the candidate is related to solving computer visual problems, machine learning).

At the same time, it is necessary to understand that the working conditions and other “buns” are known to the candidate and honestly announced in advance, otherwise not everyone will have the motivation to take the test.

1. HR & Personality Interview with Team Leader
   Personality traits that are necessary for DataScientist:
   
   • High learning It should be smart, quickly acquire new skills, be prepared and constantly develop in their field and preferably in the subject area of ​​the company.
   • Curiosity, interest in new technologies, practical experience of their use, interest in related areas.
   • Agility and perseverance - the ability for a long time to work on one problem
   • Creativity - interest in new opportunities, motivation and the ability to come up with new solutions.

How to keep AI & Data Science specialists in the company:

Here, the standard retention tools have their own characteristics.

• The opportunity to work with a guru, an expert in the AI ​​Market in Russia or other countries, the ability to write PHD, to do joint research;
• A team of strong professionals, from whom it is possible to learn and with whom it is interesting to create AI projects (Universities Top-10, employees from large companies of AI market leaders in Russia);
• The ability to write an article. To do research, and publications for international conferences (NIPS, ICLR etc.);
• Assistance in obtaining a scientific degree, including international;
• Access to primary sources.

And human values:

• Interesting tasks, the ability to make publications;
• High salary, its regular growth in accordance with the level of the market;
• Respect. Including trust expertise, recognition of achievements in the company and the scientific community (prizes, bonuses for achieving results);
• Good equipment, access to data;
• Informing about changes - employees should be aware of the company's future plans. Even in a large company, it is important to take care not to keep them in suspense;
• Care for employees - regular surveys with the possibility of obtaining honest answers. How to improve the lives of employees, to help them be more efficient (fruits in the office, musical instruments, a room for relaxation, congratulations not only on their birthday, but also on other holidays, etc.).

In conclusion, it is worth noting that it is important to know that the difference between these vacancies from the rest - the previous recruitment methods for these candidates do not work so effectively. It is important to keep a balance between the extreme shortage of specialists, the willingness to be more flexible in conditions and the need to filter and select strong professionals who can make a positive contribution to changes in business.