Renaud Allioux's tips for building the best team of data scientists

By Renaud Allioux pour Maddyness, 01/24/22

Data scientist profiles are among the most sought after by startups. But one data scientist is not like another. Renaud Allioux, co-founder of the startup Preligens, gives his tips for spotting and convincing the best talent.

As CTO of an artificial intelligence company, I spent time structuring the tech team and thinking about the profiles to hire during the different growth phases. First, at the launch of the company, defining the minimum viable product, and then during the scaling up phase and the launch of the products in production at the customers. If you are a CTO, you will need a number of data scientists, but not all of them will have the same tasks. Here are their different profiles and what I understood when I formed my AI team at Preligens, which now includes more than 70 data scientists for 200 employees.

Tip 1: A data scientist needs to code !

This is non-negotiable. This is my motto and indeed the most important thing I have learned in my 5 years as CTO. Everyone needs to be able to fit their models into a framework so that it is useful for the application and the use case. Forget about Jupyter notebooks. If you want to move beyond experimentation and deliver real AI products, your data scientists need to know how to code and deliver concrete products.

Tip 2: When starting, the first data scientist you need to hire is a “Full Stack Data Scientist”

Usually used for developers, the term full stack is also appropriate to describe a data scientist who has a wider range of skills than the greatest theoretical mathematician solving the most difficult problems with a chalk and a blackboard. A true Swiss Army knife who must have:

• strong computer skills,
• a good overall level of mathematics,
• pragmatism and boldness,
• basic DevOps (Docker, CI/CD) and visualisation skills,
• ideally 3 to 5 years of experience.
  • some more science oriented,
  • some more specialised in the iterative approach to R&D,
  • and others more focused on coding.

1. The AI production team

They are responsible for exploiting our AI stack to produce applications for clients. They are pragmatic and iterative, doing continuous R&D. They mainly have full stack or computer vision backgrounds. They do a bit of everything and are multidisciplinary. They are customer-oriented, but also good at project management and communication. It's a mix of young graduates and experienced leads who together form our "front line".

2. The AI engineering team

They are responsible for the development, improvement and maintenance of our AI stack. They are dedicated and meticulous, building the engine of the factory. They have a very high level of coding, as well as strong DS knowledge. They provide stacks for other teams to work on but also train them to be self-sufficient in adding features. They are usually more experienced than production and the best technically.

3. AI researchers

They are responsible for solving high risk, high potential problems. Although they are very scientific, they also need to understand the code to ensure that innovations are put into production. They usually have a PhD in mathematics, biophysics or astrophysics. They propose scientific exploration to solve production problems but also produce conference or journal papers. All have at least 3 to 5 years of experience.

Read the full article on Maddyness.