Deep Learning Research InternUSA (Remote)February 2021–Present
- Proposed CarneliNet, a new speech recognition model with adjustable inference resource requirements.
- Sped up training of falgship model by 30% with efficient masking and automatic mixed precision tweaks.
- Contributor to the NVIDIA's NeMo open source project for conversational AI.
- Python, PyTorch
UK (Remote)Summer 2020
- Designed and launched a pipeline to perform a continuous static code analysis of 2 Million Play Store apps that helps to drive non-SDK interface restrictions.
- Participated in Android’s Inclusive Language Fix It.
- Java, C++, MapReduce
- Increased relevance of recommendations in the internal marketing tool by inferring missing metadata of hundreds of documents with modern NLP DL approaches.
- Go, Python, TensorFlow, SQL, App Engine
- Designed and implemented a library to transform 3D data into format suitable for existing Street View Deep Learning models.
- Increased throughput of a distributed 3D rendering pipeline.
- C++, OpenGL
SWE Intern in R&D deparmentRussiaDecember 2017 – May 2018
- Implemented and compared several physically based skin deformation simulation models for 3D characters.
- Houdini, VEX, Python
- Developed a text classification model for the YouTube content rating system.
- Launched the model as a real-time production microservice.
- Python, Tensorflow, C++
- Designed experiments and implemented YouTube-scale distributed pipelines to quantify importance of graph features for YouTube language classifiers.
- C++, MapReduce, TensorFlow, SQL