Machine Learning Engineer with a passion for computer systems, and understanding how things work at a low level.
I'm currently working at Sonos as a Machine Learning Engineer in the Sonos Voice Control department.
I previously worked at ASUS Robotics & AI Center in Taiwan as a Machine Learning Engineer.
I started my career as a Data Scientist at HPS Worldwide, designing and implementing machine learning models for credit card fraud detection.
I'm fluent in French (native speaker), English, and Mandarin Chinese.
- Tract — A neural network inference engine. I have some contributions, see e.g. sonos/tract#2487, sonos/tract#2492 or sonos/tract#2515
- Performance Engineering, CPU Optimization
- GPU Programming
- Automatic Speech Recognition
- Computer Vision
| Project | Description |
|---|---|
| segcam | Real-time semantic segmentation on live camera feeds using YOLOv8-seg, with support for Metal (Apple Silicon), CUDA, and CPU backends. |
| tinygrad-tutos | Tutorials about tinygrad, an end-to-end deep learning stack. A deep dive into how modern ML frameworks work under the hood. |
| mnist-cuda | A simple CUDA-accelerated neural network for MNIST digit classification, built from scratch to understand GPU programming fundamentals. |
| tinygpt | A minimal implementation of GPT architecture using the tinygrad end-to-end deep learning framework. |
| rustynet | A neural network built from scratch in Rust, for learning — not speed. |
授人以魚不如授人以漁。— Better to teach someone to fish than to give him a fish. — Mieux vaut apprendre à quelqu'un à pêcher que de lui donner un poisson.