Scientific Software

SMT is a Python toolbox for surrogate modeling, Bayesian optimization, and design-space exploration, developed for researchers and engineers. The framework has been applied to diverse high-stakes engineering problems including rocket engine injectors, aircraft fuel consumption modeling, high-order finite element methods, solar energy planning, and wind turbine design.

Interfaces & Interoperability:

• Scikit-learn Interface: Developed interoperability layers bridging SMT surrogate models with the broader Python sklearn machine learning ecosystem.

smt-explainability is a standalone extension based on SMT dedicated to mixed-hierarchical Explainable AI (XAI). It provides automated XAI metrics, advanced global sensitivity analysis, and comprehensive visualization workflows to interpret expensive surrogate models.

smt-design-space-ext is a standalone extension based on SMT that provides robust support for complex mixed-discrete and hierarchical design variables, which are essential for framing high-dimensional system engineering problems.

smt-optim is an open-source Python package for Bayesian optimization tailored for expensive-to-evaluate black-box research applications. It is a standalone extension based on SMT that provides scalable frameworks for constrained and multi-fidelity global optimization across mixed-variable design spaces.

Key Features:

• Multi-fidelity: Implements state-of-the-art multi-fidelity frameworks (MFSEGO, VF-PI) for both nested and non-nested design spaces.
• Constraints Handling: Supports both equality and inequality black-box constraints with automated boundary management and probability of feasibility penalization.

High-performance constrained Bayesian optimization framework co-developed by ISAE-SUPAERO and ONERA. I restructured the core architecture and integrated advanced capabilities including multi-fidelity evaluation, multi-objective optimization, heterogeneous variable handling, and scalable surrogate-assisted exploration algorithms.

Framework Interoperability: Led software integrations bridging ONERA's optimization capabilities (SEGOMOE/SMT) with major aerospace design and system architecture frameworks. This includes developing interfaces for the FAST-OAD overall aircraft design framework, as well as DLR (German Aerospace Center) tools like SBArchOpt and OpenTurbofanArchitecting, enabling the resolution of realistic, hierarchical aircraft problems under hidden constraints.

GAMA is an advanced open-source platform for spatially explicit multi-agent simulation. The platform enables large-scale, GIS-integrated socio-ecological modeling using the high-level GAML language.

My work primarily focuses on the development of exploration and optimization tools integrated into the platform, as well as contributing as a regular developer to specific GAML models applied to agro-ecology.