Aleksandr Lukoianov Email & Phone Number

Helsinki, Fi

Helsinki, Uusimaa, Finland

Helsinki, Uusimaa, Finland

- Building ML and LLM-based solutions to extract structured information from webpage texts.- Developing pipelines for extracting relevant location and contact data from companies’ websites.- Leading R&D experiments including LLM-based text-to-API, universal structured information extraction, RAG-based LLM agents, multilingual strategies, efficient language identification, data augmentation, and NLP task reframing.- Collaborating with the Data Validation team and utilizing LLMs to ensure high-quality labeled data- Conducting exploratory data analysis and evaluating existing ML solutions.- Enforcing best engineering and ML practices including but not limited to code reviewing, unit/integration testing, thorough data analysis, understanding of business metrics, ML pipeline reproducibility, and model prediction explainability.- Moderating an AI reading group.Achievements:- 62.5x cost reduction and 1000x speed improvement in contact data extraction by leveraging an LLM-based solution, compared to human labelling.- +5.4% improvement in F1 exact match and 2x increase in coverage in extracting contact data using an end-to-end model with unlimited input and output context length, trained on synthetic data; this solution was 400x cheaper, 4.5x faster, and achieved +1.6% better F1 exact match compared to an LLM-based structured output extraction pipeline.- Over 20% increase in the NER model recall for location extraction through LLM-based data labelling, data augmentation, hyperparameter tuning, and optimized pre- and post-processing.- 2x increase in throughput for the location extraction pipeline by refactoring code and using efficient frameworks.

Yekaterinburg, Sverdlovsk, Russia

- Development of the intent classification system as a service, using classical heuristics approach, and shallow and deep machine learning techniques.- Building internal tools for training and validation of ML models.- Conducting experiments and A/B tests.- Developing the internal A/B testing tool for collecting, aggregating, and calculating results.- Engineering the system for regular clustering of user utterances with Ariflow to find outliers in user utterances topics.- Designing MLOps pipeline using GitLab CI/CD.- Interviewing candidates for ML/NLP positions.Achievements:- Improving the intent classifier precision by 13% with BERT fine-tuning and proper threshold strategy.- Reducing the time for decision-making from hours to minutes with new A/B testing tool.

Yekaterinburg, Sverdlovsk, Russia

- Development of the Yandex Practicum application, using Django and asyncio.- Improving service for launching and maintaining users’ Docker containers.- Designing CI/CD pipelines.Achievements:- Built CI/CD pipeline and task tracking system integration from scratch which provides a more stable and maintainable release process.- Refactoring code for service which maintains users' Docker containers resulting in uniform solution for multiple use cases.

Yekaterinburg, Sverdlovsk, Russia

Yekaterinburg, Sverdlovsk, Russia

Researching Fourier sums properties in different functional spaces. Participating in seminars.

Yekaterinburg, Sverdlovsk, Russia

Manual and automatic testing of a web application using Selenium, Postman, DevTools.

Master'S Degree, Mathematics And Computer Science, 4.8/5

Thesis: Research of the guaranteed result in differential games solved by reinforcement learning algorithms.

Diploma Of Professional Retraining, Mathematics And Computer Science, 4.7/5

Bachelor'S Degree, Mathematics, 5/5

Thesis: Order estimates for Lebesgue constants of Fourier sums in Orlicz spaces.