Responsibilities & achievements

● Implemented quite silly WebP progressive mode; ● Applying compression ratio and / or speed optimizations depending on input conditions.

Responsibilities & achievements

● Smart heater for smart home; supports WiFi, Bluetooth and ZigBee; ● Implementing WiFi-enabled board firmware using C and esp8266 rtos sdk; ● Implementing UART-based protocol for communication with heater core; ● Making draft version of client app using Flutter.

Responsibilities & achievements

● Calculates precise location of any device (or object with beacon attached); ● Schematics and firmware done on customer side; ● Implemented simple desktop client app (Python, PySide2).

Responsibilities & achievements

● Analyzing driver state, detects drowsiness, eye closure, head pose, maps facial actions to emotions; ● Implemented on Raspberry Pi, runs in 5-10 fps; ● Improving current solution by training new models (PyTorch, python, implementing C++ version on Raspberry Pi for better performance and memory footprint); ● Creating NodeJS addon to enable communication with Electron-based frontend.

Responsibilities & achievements

● Improving 3d team performance by creating blender addons (python); ● Implemented experimental real-time rendering solution using UE4 (c++, blueprints); ● Implemented side project for automatic solar panel placement using aerial images (python3, pytorch, convolutional neural networks, opencv-python, asp.net).

Responsibilities & achievements

● Optimizations applied in certain encoder modes; ● Implemented using AVX / AVX2, where it could be applied well; As a result, speed improvement gained, from 5 to 30%, depending on encoder settings and resolution.

Responsibilities & achievements

● Implemented using Torch Lua bindings and convolutional neural networks (VGG19, Resnet variants); ● Made both image and video stylization options; ● Video stylizations sped up 2x by implementing ffmpeg filter and exposing Torch C api to it.

Responsibilities & achievements

● Implemented real-time 360 degrees image stitching by transforming stream from 2 fisheye lenses; ● Integrated it into current solution as ffmpeg filter; ● On top of it, neural networks were applied to detect pedestrians; ● Made experimental SLAM version from scratch using C; ● Ported everything to Jetson TX2, made performance optimization.