Montreal-based former Lookdev Technical Director at Framestore with 8 years in film and episodic VFX. I’m transitioning into ML engineering and software development, drawn to the systems and logic behind complex production pipelines.
I've been a long-time iNaturalist user and am fascinated by plants and fungi. When I realized it's crowdsourced data could power real-world ML projects with messy observations and genuine constraints it felt natural to build on that passion.
Bachelor’s student in Computer Science at Université de Montréal.
An end-to-end deep-learning project that classifies plant phenological stages from multiple iNaturalist photos per observation.
Uses attention-based learning to reason about sets of photos rather than individual images, with swappable vision-transformer backbones (BioCLIP, BioCLIP 2, DINOv3).
Includes Hydra-configured experiments, MLflow tracking, Optuna tuning, Cleanlab label audits, and an inference pipeline that predicts from an iNaturalist URL.
PyTorch · DuckDB · Hydra · MLflow · Optuna · Cleanlab
A probabilistic ranking system that routes unresolved Québec plant observations to expert reviewers on iNaturalist after 7 days.
Achieved 3.29× lift at 94% precision on held-out 2024 data (top 5% of the test set; n = 1,374).
Uses strict temporal validation, Bayesian-shrunk observer and taxon features, and species-confusion graph features.
Includes configurable high-recall and balanced operating points.
LightGBM · scikit-learn · SHAP · DuckDB · MLflow · Optuna · DVC
Python · SQL · DuckDB · PyTorch · LightGBM · scikit-learn · Optuna · MLflow · SHAP · DVC · Git · Linux
Lookdev Technical Director, Framestore
CG asset look development on film and episodic productions including How to Train Your Dragon, Fallout SE1, Captain Marvel.
ML Engineering, Data Engineering, or Backend Engineering internships
Seeking roles where I can apply years of production pipeline experience to scalable machine learning systems.
etiennegodin@duck.com · linkedin.com/in/etiennegodin · github.com/etiennegodin