About
I’m Nicolas Chapados, a machine-learning researcher and entrepreneur based in Montréal.
One of my first jobs, as a teenager, was programming the PC version of the videogame Wrath of the Demon. In the early 1990s, that meant writing 80x86 assembly by hand and moving bytes through video memory fast enough to keep the game running smoothly. My interests eventually took me into machine learning and AI, but software remained how I put research to work. Today, studying AI systems that can write software themselves brings those interests together in a way I could hardly have imagined then.
Wrath of the Demon. I programmed the PC version in the early 1990s—a small time capsule of floppy disks, assembly code and wonderfully ambitious game-making.
Along the way, I’ve worked in speech recognition, quantitative finance, medical imaging and enterprise AI. Turning research into useful systems has often meant building a company as well as developing the underlying methods.
What I’m working on
Much of my research concerns learning and decision-making under uncertainty. How can we make useful forecasts when observations are incomplete, circumstances change and the past is an imperfect guide? How should those forecasts inform what we do next? These questions connect my work on time series, including N-BEATS and TACTiS, with my experience in investment management and business forecasting.
A curve doesn’t tell the whole story (Context is Key). Knowing that it represents solar power production changes what we can predict for the night.
More recently, I’ve been working on AI agents: how to train them to carry out increasingly complex tasks, evaluate their abilities, and make their behaviour more reliable. I’m interested in both improving what they can do and understanding when we have good reasons to trust them. That means examining how they learn, the tools and feedback they receive, and whether our evaluations capture the work we actually need them to perform.
The same request, two behaviours. In Malice in Agentland, we study how poisoned training data can teach an agent to disclose information when a trigger appears.
I pursue part of this work as co-founder and Chief Scientist of Nera Software. We’re building the system of record for verified intent, helping teams express their decisions and requirements in forms that can guide coding agents and be used to check their work.
How I got here
I studied computer engineering at McGill, then completed my master’s and doctorate in computer science at Université de Montréal under Yoshua Bengio. I worked on speech recognition and natural-language understanding at Nortel, and co-founded ApSTAT, Chapados Couture Capital and Imagia, applying machine learning to forecasting, investment management and medical imaging.
In 2016, I co-founded Element AI, where I was Chief Science Officer. Following its acquisition, I led AI research at ServiceNow from 2021 to 2025.
I’m also an Adjunct Professor of Applied Mathematics at Polytechnique Montréal and an Associate Industry Member of Mila. Research, graduate supervision and building things remain closely connected parts of my work. Sometimes an application exposes a question we don’t know how to answer yet. Those are often the questions I want to spend more time on.
You can explore my publications or find the full record in my academic CV.
Nicolas Chapados is Co-Founder and Chief Scientist of Nera, a startup building the system of record for verified intent in software systems, Adjunct Professor of Applied Mathematics at Polytechnique Montréal and Associate Industry Member of Mila. He was Vice-President of Research at ServiceNow from 2021 to 2025, and co-founded Element AI (acquired by ServiceNow in 2021), the medical-imaging company Imagia, the machine-learning firm ApSTAT and the quantitative investment manager Chapados Couture Capital. He holds a B.Eng. in Computer Engineering from McGill University, an M.Sc. and a Ph.D. in Computer Science from Université de Montréal under Yoshua Bengio, and the CFA designation. His research spans time-series forecasting, decision-making under uncertainty, healthcare AI and AI agents for enterprise work.
