At Wiremind, deep technical research is at the core of our solutions and is one of the most critical factors when it comes to staying innovative. To that end, we have in-house AI and research teams, comprised of 20+ data scientists and 5 researchers.

The team regularly publishes their research on machine learning, and we have academic partnerships with top research institutions such as the University of British Columbia and the LPSM, a probability and statistics research lab run jointly by the CNRS, Sorbonne Université, and Université Paris Cité.

A key development for our in-house research has been to host PhD students whose doctoral research is conducted here at Wiremind. These opportunities are offered to students whose research complements what we already develop, from demand forecasting for passenger transport to optimization for air cargo. Notably, our in-house causal inference framework is one result of this work, and is being continuously worked on through an ongoing PhD.

Two examples of students who are doing their doctoral research at Wiremind are Elie Lelouche and Sebastian Reboul.

I sat down with them to learn more about their research, their journeys from intern to PhD student, and how their work at Wiremind complements their academic pursuits.

In a few words, can you introduce yourself? What is your role at Wiremind?

Elie: I'm Elie Lelouche, 24, a PhD student in Machine Learning through a partnership between Sorbonne University and Wiremind. I graduated from Sorbonne last year with an MSc in Statistics and Machine Learning. My role bridges research and practical application: developing new algorithms while keeping one foot in Wiremind's day-to-day work.

Sebastian: I'm Sebastian Reboul, going into the third year of my PhD in Reinforcement Learning between Télécom SudParis and Wiremind Cargo. My role is to research how reinforcement learning can be applied to palletizing planes for the air cargo industry.

How did you first begin at Wiremind, and what made you want to stay?

Elie: My Master's program director shared a Wiremind internship listing that included the option to move into a PhD afterwards. Everything about it appealed to me: the research topic, the company, and the advisors involved. After a technical test and a few interviews, I was glad to be selected. The internship gave me my first taste of life at a fast-growing scale-up, and the atmosphere struck me. A place that is serious and highly competent while also welcoming and supportive is rare. The project was demanding and highly technical, which was exactly the kind of problem I love.

Sebastian: I saw an internship on applying reinforcement learning to the 3D bin packing problem, with a possible CIFRE grant if it went well. It was my first professional experience outside academia, and I really valued the freedom I had to explore the subject and the support from my manager. The team made it easy to stay: the average age is quite young, so everyone is relatable and easy to talk to.

What is your research focus?

Elie: My research develops forecasting algorithms that establish causal links between variables: a way to mathematically model the logical connection between events. To put it simply, traditional machine learning looks for patterns to predict what will happen next, while my work teaches a model why things happen. It is about giving a model a foundation for genuine logical reasoning. Generative modeling has shown remarkable results recently, especially in image and video, and those advances open interesting avenues for modeling causal effects. My work bridges the two: I study how they interact to design new causal forecasting algorithms, and I mathematically demonstrate that they converge to the correct answers.

Sebastian: My research focuses on applying reinforcement learning to the 3D bin packing problem, a combinatorial optimization problem that is NP-hard. What interests me most is how some actions available to an agent in one state become unavailable in the next: an item can go in a given position, but once another item is placed, that position may no longer be free.

How does your research connect to what we do at Wiremind?

Elie: My research connects directly to what we do every day. We develop demand forecasting algorithms for transportation companies, and these models need to understand true cause and effect rather than guess from patterns. They need to grasp that a price increase causes a drop in demand, and vice versa. My work helps these algorithms establish that causal link and quantify how strong and relevant the relationship is.

Sebastian: The connection is in the application. Reinforcement learning is promising in theory, but it is notoriously sample hungry: it needs far more data to learn than, say, an image recognition problem. To apply these algorithms in practice, a lot of effort goes into optimizing the code and the environment to make training more sample efficient. That is where my work comes in. We have built a Rust based simulation parallelization library that we use to train agents in Wiremind Cargo's simulators.

What does support look like day to day?

Elie: I'm fortunate to be guided by excellent academic and industrial advisors in both France and Canada. The goal of a PhD is to become autonomous, but my advisors are always there for scientific, engineering, or methodological challenges. We have weekly check-ins alongside plenty of spontaneous discussions. Wiremind provides top-tier hardware, including the GPUs and computing power we need to run experiments, and the whole team is welcoming and helpful.

Sebastian: I have three supervisors: one industrial and two academic, and I meet each of them weekly. I have real freedom over how I split my time, with a rough target of 30% in the lab and 70% at the office that flexes with the stage of the PhD. One big advantage of the CIFRE grant is access to a professional setup.

What have you learned here that a purely academic path might not have taught you?

Elie: A few things. Coding for a production environment is a unique experience you can only really get inside a company, and it is a valuable skill I enjoy developing every day. Communicating complex technical topics to teammates who are not machine learning experts has been a great exercise in clarity. And the energy here is motivating: it encourages you to stay organized, structured, and focused on real execution.

Sebastian: The biggest thing is understanding how things are actually implemented, and how hidden optimizations take a computational task from theoretical to practical. Some tasks, implemented naively, go from years of computation to a few hours. That skill is not taught in academia, and it is very useful for my research.

Where do you hope this path takes you? What does success look like a few years from now?

Elie: My immediate goal is for the PhD to lead to papers at top-tier conferences and to models deployed in Wiremind's production systems. Further ahead, success would mean continuing to tackle deeply technical machine learning problems, whether in an innovative industry setting or at a top research lab.

Sebastian: I would like to continue in academia. Success would look like postdoctoral work and, ideally, a position at a university.

What advice would you give a student wondering whether to follow a similar path?

Elie: Start by reflecting on the kind of work you genuinely enjoy. If you like tackling deeply technical problems with long-term impact, a PhD is a great route. Whether to do it in academia or industry depends on personal factors, so talk directly with PhD candidates and researchers in both worlds. And above all, be honest, kind, serious, and curious: it helps far beyond work.

Sebastian: If the CIFRE is already set up, the path is much smoother: arranging my own left a bit too much to luck. And make sure you get along with your supervisor and agree early on how you will work together over the next three years.

A Symbiotic Partnership With Real World Results

Working between academia and industry builds a rare skill set: scientific rigor on one side, having hands-on experience in a professional setting on the other. Elie and Sebastian’s experiences show what that looks like in practice, with research and industry moving together across papers and production systems, theory and real constraints, mentorship and autonomy.

For Wiremind, these partnerships keep the research that powers our products not only current with the industry, but current with the latest thinking in academia. For students, they offer a rare chance to do meaningful research with real-world impact. Together, we will continue to shape where our industries go next.

Learn more about our career opportunities on our Careers Page.