Projects

What I build

Personal projects, coursework and research — each documented with its context, technical choices and results.

How much real data does it take to detect internal defects automatically? A study run from manufacturing the parts all the way to the trained model.

A 3D optical tomography scanner: mechanics, firmware and reconstruction, end to end.

An encoder-decoder Transformer (English → French translation) written entirely by hand in C++17, trained on GPU through a home-made CUDA backend.

A convolutional neural network written entirely by hand in C++, trained on MNIST to 97.4 % accuracy — with no deep learning library at all.

A hardware implementation of ASCON authenticated encryption, the NIST standard, then full integration on a Zynq-7020 board.