The story of my life [TO READ]

My Starting Point (polished with ChatGPT)

Back in high school in France, a small group of us—the geeks, perhaps—owned an HP 48G calculator. Looking back, that little machine probably changed the course of my career.

It introduced me to Reverse Polish Notation (RPN), then to the RPL programming language, and eventually to something even more fascinating: assembly language. It was the first time I experienced what programming really meant—understanding how a machine thinks.

University was my next turning point. During my final research project at ENSEIRB (University of Bordeaux), I started programming in C, learning from Dennis Ritchie’s legendary book while implementing algorithms such as 2D FFTs and the Hough Transform.

Around the same period, during my MSc in Mechatronics, I remember implementing the Simplex algorithm directly on my HP 48G. Yes, on a calculator. I also built some of my first multilayer perceptrons (MLPs) in MATLAB as part of an Artificial Neural Networks course.

Looking back, I realize that machine learning had already entered the story—long before it became fashionable.

Around that time, I also discovered structural dynamics and, with it, a whole new way of thinking about engineering problems. The Rayleigh–Ritz method, finite element formulations, modal analysis, and numerical approximations fascinated me because they beautifully connected mathematics, physics, and computation. My bookshelf quickly filled with classics such as Vibrations des structures by Georges Venizelos, Analyse des structures par la méthode des éléments finis by Jean-François Imbert, The Finite Element Method by O. C. Zienkiewicz, and Méthode des éléments finis : Approche pratique en mécanique des structures by Michel Cazenave. Looking back, I realize how fortunate we were as students: we had the time to read these books from cover to cover, work through the derivations, and truly understand the ideas rather than simply applying the methods. Those books didn’t just teach algorithms—they shaped the way I still approach engineering today.

There was no Stack Overflow, no GitHub, and certainly no AI assistant. When you wanted to understand an algorithm, you opened a book, took a pencil, and worked through the derivation until it made sense.

Then came my PhD.

MATLAB became my everyday scientific companion for wavelet analysis, modal analysis, and vibration processing. After years of wrestling with multidimensional arrays in C, MATLAB felt almost magical: concise, expressive, and designed for scientific computing.

Cleve Moler’s books were constant companions. Years later, I had the pleasure of meeting Cleve Moler, one of MATLAB’s founders—a memorable moment for someone who had learned so much from his work.

At the same time, Stéphane Mallat’s Wavelet Tour rarely left my desk, while Pete Avitabile’s Modal Space papers became an invaluable source of practical engineering insight.

My PhD also marked my first real experience with machine learning applied to engineering. I built the database of damaged / undamaged structures with FEA, developed MATLAB implementations of MLPs for pattern recognition, and learned from Gérard Dreyfus et al (including Manuel Samuelides) excellent book Apprentissage statistique : réseaux de neurones, cartes topologiques, machines à vecteurs supports (Eyrolles, 2008).

Then, in 2006, another unexpected chapter began.

I became Manuel Samuelides’ colleague, and a few years later we started collaborating on structural optimization (with Airbus/Onera on Mixture of Experts (already GP aka Kriging). It’s funny how books that shape your education sometimes lead you to work alongside their authors. Oh Finally, Manuel also introduced me to Nathalie Bartoli at ONERA. She will become on of my strongest partner in research (on the SMT project started with Prof. Martins).

Why I Still Teach with MATLAB

People often ask me why I still teach with MATLAB, even though I also use and teach Python and Julia.

The answer has very little to do with syntax… let’s check This Post, there is a nice comparison of Matlab (Julia) and Python. who wins? Quadratic Form

One of MATLAB’s greatest strengths, in my opinion, is the quality of its documentation. Every important numerical function is carefully documented, and the implementation is almost always linked to the original scientific literature.

Take fft, for example. Its documentation doesn’t just explain how to call the function—it points you directly to the algorithms behind it:

  1. FFTW — https://www.fftw.org
  2. Frigo, M., & Johnson, S. G. FFTW: An Adaptive Software Architecture for the FFT. Proceedings of the International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 1998.

For students, this changes everything.

Software stops being a black box and becomes a gateway to the scientific literature. A single function call can lead to the original paper, then to the underlying mathematics, and eventually to a deeper understanding of the field.

I feel the same way about beautifully crafted scientific figures. A good figure can explain an idea faster than pages of equations. The MIT LAE Seminar is a wonderful example of this philosophy.

And Then…

Then came Python 2.4, during the final year of my PhD.

Then GitHub changed how we collaborate.

Then Google Colab made reproducible computing accessible to everyone.

Then ChatGPT, Claude, Gemini, AntiGravity, and the entire new generation of AI assistants fundamentally changed how we write code, explore ideas, and learn.

And the story is still being written.

I think the common thread isn’t MATLAB, C, Python, or AI.

It’s curiosity.

From an HP 48G calculator to today’s AI copilots, every new tool has been another opportunity to learn, understand a little more deeply, and pass that knowledge on to the next generation.

PS: SMT has been efficiently refactored by Claude and Remi Lafage (ONERA).

💻 “Vibe Coding” As AI assistants are transforming the way we write software, I want to explore a simple question: Can we faithfully recreate scientific software directly from published papers?

In this series, I’ll rebuild open-source implementations of classical and modern scientific methods—from optimization and surrogate modeling to computational mechanics and machine learning—using AI-assisted development. The goal isn’t merely to generate code, but to understand the underlying algorithms, assess reproducibility, and make research software more accessible to students and researchers.

Definition of Vibe Coding

Vibe coding is a software development approach where you build applications entirely by conversing with an AI in natural language. Instead of writing code line-by-line, you describe your vision or “vibe,” and AI models handle the generation, testing, and deployment, allowing you to focus purely on the creative outcome. 

Vibe coding

Recreation with my friend Claude (Anthropic)

REMAL

  • Authors: Kail Yuan, Ashwin Renganathan
  • Published: June 11, 2026
  • Link: Read Article

Summary: Multidisciplinary design analysis of coupled engineering systems requires the computation of equilibrium states in which all disciplinary coupling variables are mutually consistent. Conventional fixed-point iteration resolves this consistency problem separately at each design point, which can become expensive when disciplinary evaluations are costly and many analyses are required in outer-loop tasks such as multidisciplinary design optimization, uncertainty quantification, or digital twin updating. This paper introduces REMAL, a residual manifold surrogate modeling framework for coupled systems.

ClarkeLib

  • Author: Vadym Pasko.
  • Published: Mai, 2026
  • Link: Read Article

Summary: Tapering of a space elevator’s tether from the planet surface to the altitude of the synchronous orbit and further to the counterweight altitude can reduce the mass of the whole system, as well as provide more or less even distribution of stress along the tether rather than in the case of using uniform tether. This document presents the analyses of some tether tapering functions and shows that exponentially tapered tether is not an optimal solution for the space elevator with climbers or other point loads like transitional stations. To handle these analyses ClarkeLib – an open library for calculation of basic space elevator parameters and loads has been developed.

Recreation with my friend NotebookLM (Google)

  • Authors: J.Morlier and students
The 180 line [TO READ]

Look How Gemini ingests a full paper and is able to create an organized presentation.

Structural Optimization in 180 Lines

more mathematical details in this post QuadraticForm

Do you want to do the same ? follow HelloProf on patreon or buy me a coffee

  • Author: J.Morlier