Michał Martyniak

Engineer's diploma project made in cooperation with Maciej Rutkowski and Filip Schodowski. A lightweight convolutional neural network was trained, achieving 86% accuracy on our test set. It can identify 6 predominant instruments right now, but adding more of them would be just upgrading the same system. 1 second is enough to get an accurate classification result but 3s segments are performing a little bit better. The neural network is fed with mel-scaled spectrograms instead of raw audio.

It is capable of reading scancodes (keys), dealing with the text mode. I was able to turn on paging, protected 64-bit mode and execute Rust unsafe code (which runs even on barebone metal, without kernel libraries obviously). There is generated ISO image, which is enough to run inside QEMU + debugger (e.g. gdb). I am proud of it, but for those, who haven't experienced writing own OS firsthand, it may look crappy and straightforward. Even my lecturer wasn't 100% sure if it is feasable in Rust .

One of my very first apps, but now I know what was wrong. There is not much of Ruby code, mostly vanilla JavaScript. It has bugs

Next time I would pick:

• TypeScript
• Middleman -> static site generator
• simple_form gem
• Google Charts
• Material Design + CSS Grid instead of Bootstrap