Andrei Likhonosov Email & Phone Number

Budapest, Hu

Budapest, Hungary

- Guided the implementation and launching to production of a new version of Text Segmentation subsystem, leveraging Transformer architecture, ultimately improving segmentation quality in ABBYY Vantage.- Designed and supervised the implementation of a framework for training and inferencing neural networks written in NeoML, improving developer experience and allowing code reuse across internal projects.- Mentored automation of synthetic corpora creation for Named Entity Recognition (NER) training, this project reduced the time to support new languages and types of named entities in Vantage.- Coordinated a team of 3 developers to explore new methods to improve NER quality on tables and incomplete markup, supporting new scenarios and types of documents in Vantage.

- Implemented from scratch a subsystem for training and running NER models in production. This module has become an integral part of the NER Activity in Vantage, allowing clients to extract named entities from their documents.- Designed and implemented an automated pipeline for data preprocessing, training and evaluating of NER models, maintained related infrastructure. Established and streamlined standard processes, bringing consistency and efficiency into the project.- Developed a subsystem from scratch for training on large corpora and using word embeddings, enabling code reuse in other projects.- Conducted research on improving the quality of syntactic-semantic parser by analyzing various linguistic statistics from large corpora, applying data mining and machine learning techniques.- Developed a morphological analysis system, improving quality and performance, adding new features, simplifying the UX of internal tools. Collaborated with linguists to support new languages.

Moscow City, Russia

- Added support for new devices to the telemetry system for residential smart water and electricity meters, enabling customers to expand coverage and automate more readings.- Introduced real-time analytics dashboards and alerts using Zabbix which allowed for rapid troubleshooting.