You can view my resume in PDF format, as well as a more detailed overview of my professional experiences, below.



drawing

Computer science Research Institute of Lens

09/26 - now – Lens

PhD in explainable artificial intelligence

    I am working on explainability methods for deep learning models that are black-box in nature. The goal is to implement algorithms that provide concise, accurate, and real-time explanations for any prediction and/or model. The field of application is broad such as medicine, finance or law.

🛠️    Python, Pytorch, Scikit-Learn

📚    Neural networks, Explainable machine learning, Constraint Optimization Problem



drawing

French Institute for Research in Computer Science and Automation

03/25 - 03/26 – Rennes

Research engineer

    I worked on numerical approaches of composite manufacturing processes, modeling the flow of liquid resin through porous materials for all geometry and injection conditions. In addition to developing these numerical methods, I ensured the models were deployed and scaled to make them usable in any hardware configurations.

  • Physics-Informed Neural Network (PINNs): improve the model’s performance by incorporating a sensor-based loss function, an adaptive sampling strategy and by fixing learning issues (second-order optimizer, mixed precision, Fourier embeddings and loss balanced).
  • Finite Element Method (FEM): modify an open-source FEM solver to produce high-fidelity reference data for cross-validating deep learning models.
  • Graph-Neural Network (GNN): implemente and train an optimized model to process FEM simulation data and validate it on unseen geometric simulations.
  • Outcomes and collaborations: collaborate with energy and aerospace companies, present at major events such as AIComp and JEC World

🛠️    Python, MATLAB, TensorFlow, Pytorch, GPU computation, Gmsh, Docker

📚    Neural networks, Scientific machine learning, AI-model training



drawing

AIRBUS Operations SAS

03/24 - 09/24 – Toulouse

Intern in numerical simulations

    I built a surrogate model to predict stress outputs generated during wing-fuselage assembly phase of an A321Neo aircraft. Based on machine learning methods, the model was able to solve stresses continuous-space regression for any gap/clash scenario between wing and fuselage.

  • Benchmark: test multiple machine learning models and select the one best suited to the problem.
  • Sampling strategy: use an unsupervised machine learning approach based on the history of gap/clash values observed in production to identify relevant scenarios for simulation.
  • Data generation: automate Abaqus simulation runs of the rear spar’s detailled finite element model with Python and Shell scripts and generate efficiently training and testing data from scenario sampling.
  • Training and validation: train the surrogate model and build a full validation process to assess the performance of the model.

    The predicted stress values in the rear spar of an A321Neo for an unseen scenario were closer to the stress results obtained from finite-element method simulations. This surrogate model is now being tested on other Airbus aircraft models or other structural components.

🛠️    Abaqus Script, Python, Scikit-Learn, Shell scripts

📚    Finite-elements analysis, Machine learning, Data engineering



drawing

University of Naples Federico II

02/23 - 08/23 – Naples

Intern in aircraft design

   Attached to the Design Aircraft and Flight (DAF) research division, I worked on the hybridization parameters exploration with Monte Carlo method for hybrid-electric distributed propulsion aircraft. The aim was to provide a description of the effects of these parameters on the performance constraints, the energy consumption, and more generally on the aircraft design sizing process. The large number of possible combinations means that sizing algorithms need to be coupled with statistical computation method such as Monte-Carlo method.

  • OpenVSP simulation: demonstrate the improved aerodynamic performance of a distributed propulsion aircraft during takeoff and climb.
  • MATLAB coding: code a sizing algorithm based on the works of Delft University of Technology for an hydrid ATR 72-600 aircraft.
  • Monte-Carlo computation: generate thousands of aircraft designs with random parameter values.

    I concluded that the set of convergent designs is located in a delimited space. By choosing precise hybridization parameter values for each flight phase, I designed a draft version of an optimal hybrid aircraft with the lowest takeoff weight, saving between 6.5% and 16% in kerosene in a short term configuration of battery technology scenario. My draft design and results were compared with DAF’s work, which helped reinforce confidence in their high-fidelity methods and aircraft sizing models. During this internship, I communicated with my team in Italian and wrote in English.

🛠️    MATLAB, Git, LaTex, OpenVSP

📚    Aircraft sizing algorithm, Flight mechanics, Aerodynamics



drawing

ELISA Aerospace

09/19 - 09/24 – Saint-Quentin

School of aerospace engineering

  • Degree: with honors mention given by the exam board.

🛠️    C++, Abaqus, Ansys Fluent

📚  Computational science, Statistics and Optimization, Flight Mechanics and Aerodynamics, Finite Element Method, EEA (Electrical Engineering, Energy, and Automation)