Internship - Software Development For a Bayesian Optimization Framework H/F CEA

At the Institute of Applied Sciences and Simulation for Low-Carbon Energies (ISAS) of the CEA, we operate at the interface of theoretical physics, applied mathematics, and computer science. Research at the institute involves the development of computational tools to model complex systems in material science. Bridging the gap between academic prototypes and robust software is essential to translate these models into reliable solutions for applied research.
One specific application concerns additive manufacturing, which enables the creation of metallic components for industrial use. These parts involve complex geometries which need to be engineered to ensure the desired physical and mechanical properties. Designing these structures requires navigating a high-dimensional parameter space to reach arbitrary optimal configurations under constraints. To address this, the Laboratory of Artificial Intelligence and Data Science (LIAD) and the Laboratory of Engineering of Surfaces and Lasers (LISL) have jointly developed a prototype software suite. This tool couples finite element analysis with constrained Bayesian optimisation to automatically identify optimal geometric parameters.
The current Python codebase provides a functional and documented implementation of the mathematical framework, successfully validated on specific use-cases. To extend its utility to broader additive manufacturing applications, the software requires architectural modularisation. The objective is to transform this specialised implementation into a generic, installable library capable of accommodating diverse geometrical constraints and physical models.
The intern will undertake the following tasks, which will provide the opportunity to learn and apply the principles of software engineering best practices, specifically in the areas of architecture, standardisation, and packaging:
Refactoring and Standardisation: Unify the Command Line Interfaces, implement robust path handling, and standardise argument parsing to abide by Unix/POSIX standards.
Architectural Decoupling: Separate the core optimisation engine from simulation-specific parameters. This involves building a modular interface to load user-defined constraints and physical models, allowing the package to accommodate new geometries without modifying the internal codebase.
Robustness and Validation: Implement strict input validation (e.g. using Pydantic or JSON Schema) to prevent runtime errors, and develop a suite of unit and integration tests integrated into a continuous integration (CI/CD) pipeline.
Documentation and Packaging: Write comprehensive developer and user documentation (Sphinx/MkDocs), create tutorials, and finalise the Python package configuration for standard installation.