Day in the Life of a Lead Machine Learning Engineer at Wiremind: An Interview with Mathilde Bleu

Sruthi Kolukuluri

4 September 2024

Cargo

Share

Being a Lead Machine Learning (ML) Engineer is not an easy feat — especially at a company like Wiremind where ML is a crucial element of our solutions. It requires both technical prowess and managerial expertise to meet the constantly evolving customer demands.

We spoke with Mathilde Bleu, our Lead Machine Learning Engineer, to get a glimpse of the inner workings of the ML team. Discover what led her to her current role, Wiremind’s tech stack, and how her team is working to meet clients’ needs.

Tell us a bit about your background and what led to your current role as the Lead Machine Learning Engineer

For as long as I can remember, I’ve been intrigued by the application of mathematics and how it can be used to solve real-life problems. This curiosity led me to pursue my Engineering degree at ENSTA, followed by a Master’s in Mathematical Modelling at UCL. It was during my internship at the IBM research lab in the Netherlands that I transitioned from Mathematical Modelling to ML and started my journey in this field.

Initially, I worked at a data consultancy firm, delivering ML projects to multiple clients spanning from the luxury industry to national telecommunication companies. However, it was my interest in working for a fast-growing company that pushed me toward Wiremind.

I joined Wiremind as the Lead ML Engineer in 2022. Apart from Wiremind being a fast-growing company, this move was partly also motivated by the fact that ML is an important part of the products at Wiremind, both in terms of added value and in resources dedicated to it. Starting from data engineers to ML engineers and optimization engineers — several teams work together on the ML and optimization components. Extra care is given to the quality of code, performance of the models (both in terms of metrics and inference time), reliability of the ML modules, and continuous improvements of our pipeline and current process. This kind of work culture inspires me and pushes me to be my best version.

What does your typical day look like?

As the Lead ML Engineer, my days are split between:

• Technical tasks such as analyzing client data, training ML models, writing new components for our ML pipelines, and reviewing code written by team members.
• Tasks related to the current and future usage of our models by the clients such as following up on the implementation, ML performance monitoring, and brainstorming with the product team on future use cases.
• Management tasks like coordinating between teams and gathering the right information to execute our tasks perfectly.

Walk us through a day in your team. What tasks do you execute on a daily basis?

Our work starts when the client chooses our solution – in the implementation phase. Depending on the phases of a client implementation, we are mostly executing typical ML tasks such as:

• Analyzing data and iterating together with our product team and the client to get a better understanding of data patterns, outliers that need to be removed, and identifying important features like what are the most important characteristics of a flight/booking, which ones are most useful to help forecast a behavior, etc.
• Working on creating new features, adjusting existing ML models, and/or defining new ML models to solve our problems.
• Training and evaluating models, and iterating them together with the product team.
• Monitoring deployed models.

Besides these typical ML tasks, we also work on constantly improving our ML stack in order to iterate more quickly, have better feature engineering, prevent technical debt, and improve our monitoring tools. Additionally, we have regular internal meetings to share knowledge and find solutions together, or to take advantage of a technique that has proven to be efficient for another product of Wiremind.

What is your favorite part about working at Wiremind?

My work at Wiremind changes depending on the client and the implementation phase for our products. For example:

When implementing a capacity forecast model, we get to anticipate the best final available capacity taking into consideration the baggage data (like how much baggage will be carried by the passengers on a flight, depending on the route and/or season) and also the fuel consumption data (like how much fuel will the aircraft need for takeoff and how much fuel will be actually consumed).

We also get to explore and observe different types of booking products, different booking flows, and many more items, varying from one client to another based on their respective market and business context.

The best part about this is that the number of things we get to learn will keep expanding as we get more clients under our portfolio. So, it’s very exciting!

Which of the products in the CARGOSTACK utilize ML?

ML is implemented across our CARGOSTACK Optimiser modules, which aim at solving multiple customer pain points. The solution provides decision insights across capacity, revenue, and pricing management functions, optimizing the full air cargo cycle.

We are currently applying ML across several features of the CARGOSTACK:

• Demand forecast feature that allows customers to maximize revenue
• Overbooking forecast feature that enables clients to maximize load factors and avoid empty positions
• Capacity forecast feature to anticipate the actual capacity of the flight when they are departing.

We leverage customers’ historical data and train our ML models to put the data to work and generate recommendations for various commercial processes, improving accuracy.

What framework/stack are you on and how did you decide on it?

At Wiremind, our ML teams use a shared stack that we’ve meticulously developed to meet our needs. This stack is constantly evolving and improving. It is based both on open-source tools (for instance pipeline orchestrators, experiment trackers such as MLFlow, etc.) and in-house tools and ML libraries.

Our main focus is on building an internal framework that enables ML engineers and remove as much blockers as possible.

For this we have created our own library of customizable pipeline components crafted specifically for the needs of ML for revenue management, that is shared between the teams, and avoid each individual developing their own pipeline from scratch for each new client or project. We also work on the efficiency of our workflows by carefully deciding on the tools to include and the computing resources allocated to them, so the team can iterate quickly. Additionally, we continuously automate all repetitive and error-prone tasks (such as model deployment) to increase the reliability of our workflows.

Do we have any research partnerships? Have you published papers based on your work at Wiremind?

Our internal team is working on several research partnerships with top-tier universities in Europe. We have a partnership with the research labs of Telecom SudParis, with whom we are co-managing a PhD on reinforcement learning.

We are also collaborating with Sorbonne University (Paris) and the University of British Columbia (Vancouver) on research topics. We’re tackling the same initial problem with both, but each is exploring a different path. Our main goal is to forecast elastic demand matrixes in our Revenue Management Solution CAYZN using deep learning models and submit a paper by the end of September 2024.

As a young manager in your field, how do you maintain a work-life balance?

In my time at Wiremind so far, I’ve had the chance to recruit and mentor interns who turned into full-time employees and juniors who have developed significant skills in the field. I try to maintain a work-life balance by taking part in multiple activities outside of work. I enjoy running and try to attend marathons in Paris. I take some runs during lunch breaks at the office, thanks to our flexible work schedule to unwind when needed. I also encourage my colleagues at Wiremind to go on runs and organize races with them. It’s been a great experience so far!

Want to be a part of our ML team at Wiremind? Work on our industry-leading solutions spanning across passenger transportation, air cargo, and entertainment industries. Apply now.