Athmane BenarousAthmane Benarous
I make machine learning hold up on real equipment.
My work spans slow-leak detection for high-voltage circuit breakers, ML pipelines over 600K+ materials, and LLM agents a food plant uses every day.
Open to ML, applied AI and software engineering roles
- M.Sc. Computer Science (Machine Learning), McGill, December 2026
- Based in Montréal
- Canada or US (TN-eligible)
- In person or remote, open to relocation
Selected work
6 case studies
Slow-leak detection for high-voltage breakers
Hydro-Québec Research Institute (IREQ) · May – Sep 2026A detector for slow SF₆ loss in 145–315 kV circuit breakers, with an evaluation protocol built for overlapping time-series windows. Not certified for operational use.
0.867 vs 0.733sensitivity vs the physical baseline on synthetic leaks, at an equal false-event budget; the gain is conditionalLLM agents in production at a food plant
Russet House (food processing) · Jan 2026 – presentLLM agents that plant staff use every day to get answers from the company's ERP data, SharePoint documents and Teams meetings, built on Copilot Studio.
Daily useby plant staff, in production since January 2026Self-learning line-speed advisory
Russet House (food processing) · Jun 2026 – presentProposes a line speed to operators from live SCADA data, learns online with recursive least squares, and takes their feedback through a chat interface.
1 recommendationfor the one input speed operators control; they accept, modify or refuse itML pipelines over 600K+ materials
Chemia Discovery · Feb – Aug 2026Azure ML pipelines that train, evaluate and backfill CGCNN property predictions across 600K+ materials, with an evaluation layer and a training dashboard.
30+ h → < 1 hpipeline runtime, after parallel S3 fetching and batched API calls3D neuron soma segmentation platform
McGill capstone, team of 4 · Sep 2024 – present (research ongoing)An open-source Vaa3D plugin that turns one marker per soma into a 3D segmentation, a shape model and synthetic training data. My part: the segmentation module and its evaluation.
1 markerper soma (center and radius) to get its full 3D binary maskAlphaGateau for Breakthrough
McGill, team of 3 · May 2026AlphaZero-style agents in JAX on Breakthrough, comparing a graph-attention encoder with a CNN and testing whether 5×5 pretraining transfers to 8×8.
17% → 67%GNN win rate at iteration 5 after 5×5 pretraining (12-game greedy evaluation)
Experience
Software Developer
Oct 2023 – PresentRusset House (food processing) · part-timeBuild a self-learning line-speed advisory and LLM agents the plant uses every day, and maintain its Ignition SCADA, SQL Server and Power BI systems.
Machine Learning Research Intern
May – Sep 2026Hydro-Québec Research Institute (IREQ) · Mitacs internship, with McGill and MilaBuilt and evaluated a detector for slow SF₆ loss in 145–315 kV circuit breakers from quarter-hourly pressure and temperature records.
Machine Learning Intern
Feb – Aug 2026Chemia DiscoveryBuilt Azure ML pipelines that train, evaluate and backfill CGCNN property predictions across 600K+ materials.
ML / Cybersecurity Intern
May – Aug 2025Hydro-QuébecBuilt an NLP chatbot (naïve Bayes) that navigates a knowledge graph of criticality indicators.
Cybersecurity Intern
May – Aug 2024Hydro-QuébecAutomated cybersecurity reporting with Python, Power BI and Power Automate.
CRM Developer Intern
May – Aug 2023Russet HouseHelped gather requirements, designed an SQL database and modernized web pages with ASP.NET and C#.
Business Intelligence Intern
May – Aug 2022CIUSSS du Centre-Sud-de-l'Île-de-MontréalBuilt Power Apps, Power BI and Power Automate solutions and TypeScript web components in an Agile team.
Education
M.Sc. Computer Science (Machine Learning)
Sep 2025 – Dec 2026McGill UniversityB.Eng. Software Engineering (Co-op), Minor in Aerospace Engineering
Sep 2021 – Aug 2025McGill University
Relevant coursework (9)
- Reinforcement Learning
- Applied Machine Learning
- Computer Vision
- Natural Language Processing
- Intelligent Software Systems
- Distributed Systems
- Parallel Computing
- Network Science
- Machine Learning Applied to Climate Change
Skills
- ML
- PyTorch
- JAX
- scikit-learn
- TensorFlow
- NumPy
- pandas
- GNNs
- Time series and anomaly detection
- Leakage-safe evaluation
- RL (MCTS)
- LLM agents
- Data and infrastructure
- Azure ML
- AWS S3
- Slurm GPU clusters
- SQL Server
- PostgreSQL
- MongoDB
- Git
- Software
- Python
- SQL
- C
- C++
- Java
- TypeScript
- JavaScript
- C#
- Bash
- React
- Next.js
- NestJS
- Spring
- Industrial
- Ignition SCADA
- Modbus/OPC
- Power BI
- Power Automate
- Power Apps
- Languages
- French
- English
- Arabic
Research & papers
A Platform for the Semi-automatic Segmentation, Characteristic Shape Analysis and Simulation of Neuronal Somas
Sep 2024 – presentA Vaa3D platform to segment, model and simulate neuronal somas in 3D light-sheet brain images.
AlphaZero-style self-play in JAX, comparing a graph-attention encoder with a CNN on Breakthrough.
Where the training energy of a ViT-Huge video Transformer goes on one GPU, phase by phase.
A multi-view graph benchmark for stance detection in decentralized social networks.
A meta-model that predicts when fine-tuning a global LSTM flood model pays off for a river basin.
How the age of training data affects sentiment-model performance on Amazon reviews.
Projects
6 shown
DeepMonitor
Since Sep 2025Co-founder · Early prototype (breadboard testing), on holdVibration sensing for food-grade industrial machinery. A battery-powered ESP32 sensor extracts FFT and statistical features and runs a small autoencoder on-device that learns each machine's normal signature, with a self-calibrating alarm threshold; a local server and dashboard collect alerts.
BargainBot
Jan 2025McHacks 12An AI agent that finds internet deals and negotiates on the user's behalf.
GreenFilter AI
Jan 2024Harvard AI EarthHack · Top 15%A layered LSTM model that validates sustainability problems and their proposed solutions.
Purposeful
Apr 2023McGill, team projectFull-stack idea-sharing platform built as a team: Spring backend, Next.js frontend, Gherkin acceptance tests.
Museum Management System
Dec 2022McGill, team projectSystem to manage a museum's collection, staff and visitors: Spring backend, Vue.js frontend, Gherkin acceptance tests.
Flight computer simulation
2022 – 2023McGill Rocket TeamSimulation backend for the rocket's flight computer: RocketPy flight dynamics integrated with STM32-based hardware.
Earlier hackathons
Higher, Higher!
Jun 2022SimpliHacksAn interactive React game about environmental issues, prototyped in Adobe XD.
Akademos
Feb 2022Code.Jam() XIAndroid app for students with to-do lists, reminders and study timers.
Hermes Discord Bot
Jan 2022McHacks 9Discord bot for managing and delivering messages.