Edouard Belval’s passion for programming dates back as far as he can remember. At age 11, he became immersed in LEGO Mindstorms home robotics kits and, as a teenager in Quebec, cut his teeth on C++ before completing a three-year computer science CEGEP (Collège d’enseignement général et professionnel) program after high school (CEGEP is an intermediate level of education specific to the province).
“There I was introduced to the concept of machine learning by one of my teachers,” Belval said. “I got an internship to build a document classification system for CEGEP. From that project on, I realized that’s what I wanted to do as a career.”
Belval entered École de Technologie Supérieure, an engineering school in Montreal, to study software engineering and eventually completed an internship with Element AI in 2019. Around that time, an AWS senior applied scientist, Yuting Zhang, found him on the GitHub developer network, where a few of his projects were gaining attention.
“They said my profile was a good fit for their team and the work they were doing, and asked me to interview for an internship,” Belval recalled. “I knew I’d be able to do machine learning there, which was my goal.”
He landed a software development internship with the AWS AI team to support the Textract service. Due to COVID-19, the Pasadena-based internship turned out to be a remote assignment. His first project focused on extracting tabular data from customer-submitted PDFs, and Belval helped develop a cost-saving technology to optimize the PDF conversion workflow process.
“Because I’m based in Montreal, I was collaborating both with people on the West Coast and in Israel because I was right in the middle of their time zones,” he said. “I could catch scientists in Israel later in their day and scientists on the West Coast at the start of theirs, which was great.”
An extended opportunity
Belval’s four-month internship was extended another four months while he finished his undergraduate degree. In this second phase, his work contributed to reducing a single model’s processing time by 120 milliseconds which, when put into production at Amazon’s scale, results in significant efficiency gains.
“It was quite a challenge because of the sheer size and scale — which is a common theme at Amazon,” he said. “What I found very stimulating is the fact that work like this, which might seem small, is something other smaller companies wouldn’t explore. But the impact we are able to drive because of our scale is significant.”
Throughout his internship, Belval had weekly meetings with his managers to monitor progress against his goals. He also jumped at the chance to collaborate and learn from seasoned AWS scientists, who challenged and inspired him daily.
“When I started, I remember telling my friends at university that everyone I was working with was much, much smarter than me,” Belval said. “And that’s an awesome feeling, because an internship is an opportunity to learn and tap into all of these experienced professionals who can push you much further than you’d be able to go on your own.”
Joining the documentation-centric Amazon culture helped Belval hone his writing skills as well.
“There’s a saying in data science that if you don’t write down your findings, then it’s as if you never did the experiment at all,” he said. “Before my internship, I didn’t have a routine of journaling everything and would often have to redo experiments. Now, I write everything down. That’s probably been my biggest learning.”
In January 2021, he was offered a role as a research engineer. After eight months in the role full time, he transitioned to part time while pursuing a master’s degree at Université de Montréal.
Everyone is encouraged to push ideas, even as an intern. You are very rarely limited in what is available to you to research or explore.
“Edouard is an extremely talented engineer,” said Thomas Delteil, AWS senior applied scientist and Edouard’s current manager. “As an intern, he became the subject-matter expert on several of our developing technologies unrelated to his own project because he had a knack for jumping into complex problems and helping the team out. Edouard embodies the Amazon leadership principles. He’s customer obsessed, has a strong sense of ownership and bias for action, and he delivers bar-raising results.”
“I’m working on a Visual Document Understanding project that might pan out to a published paper, which is always something I’ve wanted to do,” Belval said. “I’ll be doing more work in deep learning model optimizations to make AWS AI algorithms faster, which is very interesting to me because there is a real quantifiable impact to the work — not only by improving the customer experience by delivering answers to queries faster, but also by helping reduce expenses on the business side.”
Advice for fellow interns
When asked what advice he’d give to fellow data scientists considering an internship with AWS, he offered the following: “Be clear with your manager on your expectations, goals and definitions of success, and document everything. Also, don’t be afraid to accept that your goals may change during your internship. Internships can start with smaller projects, like my PDF extraction project, but once I showed I could achieve results and was fast in my execution, they quickly reoriented me to a different project. Also, interns are often assigned ‘moonshot’ projects with uncertain outcomes, so it’s important to be at peace with the fact that you may fail.”
AWS internships are, in many cases, defined and dictated by the interns themselves.
“AWS is a great lab environment,” said Belval, who will complete his master’s coursework in December before starting a six-month lab assignment with AWS. “We have a ton of resources, and everyone is encouraged to push ideas, even as an intern. You are very rarely limited in what is available to you to research or explore.”
