Bellatrex
Explains individual Random Forest predictions using a small, diverse set of representative rules. The method supports classification, regression, multi-label, multi-target and survival settings.
Selected work
These projects show how I move between method development, evaluation and usable software. My current enterprise work is not public, so the open-source projects below provide the clearest inspectable examples of my technical approach.
IntervalSHAP
Extends SHAP-style explanations to time-to-event predictions, focusing on feature importance over selected time intervals.
Active Learning for Survival Analysis
Code and experimental material for active-learning strategies with censored outcomes and incrementally disclosed labels.
EDGEHOG
A workflow based on Histograms of Oriented Gradients for measuring fibre alignment and directionality dispersion in biological images.
Applied industry work
At CGI SmartLab, I work on an enterprise RAG application and contribute to prompt behaviour, retrieval, automated LLM evaluation and front-end functionality. At Predikt.ai, I worked on time-series forecasting, model validation and uncertainty estimation for finance use cases. These projects are described at a non-confidential level on the homepage and in my CV.
Research record
My PhD dissertation, Supporting adoption of AI in Survival Analysis, was supervised by Prof. Celine Vens. The full dissertation and Google Scholar profile provide the complete academic record.