Backend Engineer/Data Scientist
• Projects: Smart ID Engine (IDs scanning SDK for over 2484 documents worldwide), Smart Document Engine (high-performance documents' analysis and data extraction software)• Job responsibilities: - Training, validation and optimization of Deep Learning models for Computer Vision tasks- Implementation of custom cost functions and layers using CUDA- Integration of trained models into backend components of the system- Selection and verification of appropriate quality measurement for trained models - Implementation and enhancement of synthetic data preparation tools- Gathering datasets and their adaption for trained models' verification- Development of data visualization, labeling transition, automarkup and quality measurement tools for trained models, implementation of automated data preparation scripts- Establishing communication between various company's departments, coordinating and controlling all stages of task's solution- Software Profiling and Refactoring- Development of engine's core components- Research of software optimization possibilities and their quality assurance - Experiments with state-of-the-art approaches in various Computer Vision and Image Processing areas• Achievements:- Built specialized models for Text Detection and Segmentation, integrated them into the software product- Optimized hyperparameters for several models' post-processing resulted in a weighted average score increase by 14%- Decreased number of false positive cases in document forensics task by 5x- Implemented custom cost layer based on a concept of Wasserstein metric - Prepared a comprehensive benchmark for Text Detection models' verification- Organized Text Detection model's pipeline, from data preparation to quality measurement- Accelerated recognition subsystem by 4x times- Implemented effective serialization and deserialization of one of the key structures
