Yacine Mahdid

Yacine Mahdid is the CEO of Datadom, where his team works alongside engineers in software, finance and manufacturing to test whether machine learning improves a product they already run. Together they define what would count as an improvement and build a benchmark from the customer’s own data. For a manufacturer assessing weld quality, a test case could pair a weld monitor’s trace with a photograph of the same weld. The aim is for the customer’s engineers to be able to run the model, its data pipeline and the underlying infrastructure themselves.

In a February 2025 essay, Yacine describes building applications around trained models as stochastic software development. He distinguishes training, which changes a model’s weights, from the work of an application engineer using that model. The engineer chooses what input to send and has to decide how the program should behave when the same input yields different answers.

Building Axya while studying brain networks

In 2019, Yacine co-founded GRAD4, later renamed Axya, with Félix Bélisle-Dockrill and Nicolas Gauthier. At Centech in Montreal, the founders tested whether they could help manufacturers find suppliers in Quebec to take on subcontracted work. In the winter of 2021–22, Yacine was both Axya’s chief operating officer and a doctoral student in McGill’s Biosignal Interaction and Personhood Technology Lab. He had earned his BSc at McGill in 2017.

Yacine coauthored a study of brain network motifs that followed nine healthy volunteers as they lost and regained responsiveness under anesthesia. Using 128-channel EEG, the team isolated activity in the alpha frequency band and estimated the direction of connections between recording sites. Examining three sites at a time let the researchers track both how often a pattern of connections, or motif, appeared and where it occurred.

The counts did not distinguish responsive from unresponsive states. The locations of two motifs, however, shifted as volunteers lost and regained responsiveness. One motif returned to its baseline distribution across recording sites at least ten minutes before the volunteers began responding again.

Teaching machine learning through projects

When teaching machine learning to undergraduates at McGill, Yacine encouraged students to start with a project like one they wanted to build, then make a small version of their own within a month. That first attempt gave students a way to choose what to study next. They could identify one part of their project they did not understand and return to the relevant math or algorithm with a specific question.

Yacine’s EEG headset converter gives nearest-neighbour search a practical job. Headsets can have different electrode layouts. To map channels between them, his MATLAB code reads two tables of electrode positions and finds 20 nearby reference channels for each position to be mapped. He implemented a rule based on distance alone and another that gives matching channel labels priority. He selected the label-first rule for this experiment.

Yacine also publishes machine-learning video tutorials and regularly reviews research papers in learning with yacine, his newsletter on deep learning and the brain.