Victor Kovalev

ML/NLP Engineer

MTS Lab

Builds production LLM and NLP systems for database metadata, including classification, evaluation, knowledge extraction, and model serving.

• Develop Data Scout, an AI service for automatically analyzing and describing schemas, tables, and columns in corporate databases. • Designed an LLM pipeline for generating descriptions of tabular data: analyzed the contribution of different context sources, changed the input data, compared models and prompts, and improved the final description quality. • Took binary and multiclass NLP classifiers of data objects to production, classifying them by usage categories and industry standards; compared BERT-based models, zero-/few-shot LLM approaches, and classical ML baselines.

Research Engineer / Data Scientist

Innopolis University

Conducts bioacoustics research, including self-supervised audio pretraining and a real-time sperm whale click detector published at ACM Multimedia 2025.

• Lead research projects analyzing marine mammal audio: formulate hypotheses, collect and clean large collections of hydrophone recordings, prepare datasets, and conduct downstream evaluation. • Developed self-supervised pretraining approaches for general bioacoustic tasks: trained models on unlabeled audio and evaluated learned representations on downstream tasks. • Developed a real-time pipeline for detecting sperm whale clicks using self-supervised embeddings and a BiLSTM; results were published at ACM Multimedia 2025.

ML Engineer Intern

MTS Lab

Explored graph-based and other ML methods for mobile network coverage optimization and prototyped an internal text-to-SQL agent.

• Researched classical ML, GNNs, and basic RL approaches for upgrading a mobile-network coverage optimization algorithm; assessed their suitability given the network's graph structure. • Built a reproducible data preparation, feature engineering, and GNN training pipeline using mobile-network topology to predict potentially problematic areas. • Developed a prototype internal text-to-SQL agent using an LLM to run analytical queries against data in a secure environment.

Data Scientist

VF-Inno

Built and productionized text classification and NER models for a news service, and integrated an image-upscaling approach.

• Developed ML components for an autonomous news service aggregating data from multiple sources; contributed to data preparation, model training, and integration of solutions into the product pipeline. • Fine-tuned and compared models for binary and multiclass text classification, including BERT-based models, zero-/few-shot LLM approaches, and classical baselines; built training/evaluation pipelines and moved selected models into production. • Developed an NER pipeline for Russian-language news using Natasha and LLMs; used extracted entities to build and analyze a knowledge graph.

Bachelor's degree, Informatics and Computer Engineering

Innopolis University