Dmitriy Stoyanov Email & Phone Number

Russia

Санкт-Петербург, Россия

- Fine-tuned YaGPT model for the task of product generation, slashing the defect rate by 2.4 times, which enhanced the accuracy of product information presented to customers. Additionally, optimised latency by 1.5 times through quantization, contributing to a smoother user interface experience- Created a model for semantics change detection based on BERT architecture. Model finds thousands of bad matches with a precision of 50% on a daily basis- Drafted a tool for finding trending e-com… Show more - Fine-tuned YaGPT model for the task of product generation, slashing the defect rate by 2.4 times, which enhanced the accuracy of product information presented to customers. Additionally, optimised latency by 1.5 times through quantization, contributing to a smoother user interface experience- Created a model for semantics change detection based on BERT architecture. Model finds thousands of bad matches with a precision of 50% on a daily basis- Drafted a tool for finding trending e-com topics based on search sessions Show less

Санкт-Петербург, Санкт-Петербург, Россия

Development and analytics in Information Retrieval team- Trained and implemented a novel HNSW and Catboost-based offer classifier, confidently classifying 30% of new offers, resulted in reduced on-boarding time by over 41%- Decreased the flow of customer complaints by creating an anomaly detection model with a precision of 30%- Adapted an existing ML service, based on map-reduce solution for use in run-time, which reduced latency for predictions and freed up the capacity of… Show more Development and analytics in Information Retrieval team- Trained and implemented a novel HNSW and Catboost-based offer classifier, confidently classifying 30% of new offers, resulted in reduced on-boarding time by over 41%- Decreased the flow of customer complaints by creating an anomaly detection model with a precision of 30%- Adapted an existing ML service, based on map-reduce solution for use in run-time, which reduced latency for predictions and freed up the capacity of operators- Improved the algorithm of matching to SKU and validated it by performing B2B tests. As a result, boosted precision from 90% to 95%- Applied the sharding mechanism in several services to achieve horizontal scaling- Automated statistics calculation and built a dashboard to analyse problems with matching- Optimised total CPU time for predictions in 2 times by profiling map-reduce service and excluding unnecessary computations Show less

Санкт-Петербург, Санкт-Петербург, Россия

Development in Cloud Billing team- Examined existing BPMN solutions and formulated a concept for developing multi-tenant service using Camunda, focusing on evolvability, maintainability and fault tolerance- Originated a telegram bot based based on Kotlin for automating and verifying the process of creating release branches, which simplified developers duties and reduced time for release branch preparation from 45 minutes to 15 minutes- Performed load testing and searched for bottle… Show more Development in Cloud Billing team- Examined existing BPMN solutions and formulated a concept for developing multi-tenant service using Camunda, focusing on evolvability, maintainability and fault tolerance- Originated a telegram bot based based on Kotlin for automating and verifying the process of creating release branches, which simplified developers duties and reduced time for release branch preparation from 45 minutes to 15 minutes- Performed load testing and searched for bottle necks- Ensured restart-safe features for cloud services by using idempotent updates, Kafka transactions Show less

St Petersburg City, Russia

- Designed data model and created micro-services using SpringBoot, PostgreSQL, MongoDB, Kafka, Swagger- Participated in demonstrations with German customers

St Petersburg City, Russia

- Implemented a web application for online assessments on the Joker conference with 100+ active users using Spring Boot, Angular, MongoDB- Extended base assessment functionality with code tasks, using Docker for evaluating users submission code

Магистр, Технологии Промышленного Программирования

Бакалавр, Прикладная Математика И Информатика