- Lead & maintainer of ConvAI module of Transfer Learning Toolkit (ASR & NLP & TTS)- Implemented YOLOv4 detector pipeline- Realistic 3D scene simulation using NVIDIA Omniverse- Transfer Learning Toolkit: we are continuously optimizing our product, check it at the link below.

Deep learning for video analysis (IVA)- Designed & implemented an in-house simulator (UE4 based) that automatically generates simulated images and object labels. This enables us to expand training dataset with much less labor effort and legal risk.- Main developer of Transfer Learning Toolkit (see link below)-- Implemented TensorRT custom plugins for multiple network architectures.-- Implemented UFF exporter for multiple network architectures for model deployment.-- Implemented SSD detector pipeline (including the detector itself, training & inference pipe and model pruning)-- Implemented YOLOv3 detector pipeline (including the detector itself, training & inference pipe and model pruning)-- Implemented SqueezeNet template for both classification and detection.- Implemented part of (deep learning) face feature registration pipeline for company access control.

Deep learning for video analysis (IVA)- Designed & implemented a single neural network to predict multiple face attributes (age, hair color, race etc.)- Designed & tested multiple features / networks for user-unattended spoof (face liveness) detection.

Worked on EOS (Linux-based operating system) kernel- iproute package- dmesg dump- rewrote some kernel patches for new version of Linux kernel