Hey there! 👋

My name is Kanstantsin Sokal and this is my blog about software engineering.

Born and raised in Minsk, Belarus   Currently residing in San Francisco, CA

I build cross-platform core libraries for On-Device ML and AR at Google. Those libraries drive face AR features inside YouTube, Meet, Duo and Arts & Culture apps on Android, iOS, Web and Linux servers. Parts of my work are publically available through MediaPipe and ARCore.

Before joining Google, I worked in a tech startup AIMatter. We built Fabby, the first app to ever run a convolutional NN on a smartphone in real-time and solve the background segmentation task. The company was acquired by Google in August 2017. The underlying tech is publically available through MediaPipe, ML Kit and TensorFlow Lite.

Before joining AIMatter, I had a long career in competitive programming. I won a silver medal at the IOI 2013 and reached the ACM ICPC World Finals 2016. I’m a red coder on CodeForces and TopCoder, however it’s been quite a while since my last contest.

I like pragmatic solutions that get things done! 🛠

I write low-level C++ code to tackle various problems in ML and AR. That includes turned raw ML solution outputs into useful AR geometry primitives using popular libraries for CV and linalg (OpenCV, Eigen, MathFu). I generally work with older graphics APIs (OpenGL 4.3, OpenGL ES 3.0, WebGL 2.0) as they offer great device coverage. I write high-level TypeScript/JavaScript code and run it inside an embedded JavaScript runtime (JavaScriptCore, V8, QuickJS) across many platforms.

In my free time, I like brewing coffee ☕️, cooking 👨‍🍳, running 🏃, hiking 🥾, reading fiction 📖, watching movies 🍿, playing video games 🎮 and spending time with my adorable doggie Mia 🐶!