Dmitry Balobin

Senior ML Engineer, Alice on Search

• Optimised LLM training infrastructure on 32–64 GPUs, making training more than 2× faster (huggingface: yandex/AliceAI-T5-35B-A0.6Bhuggingface: yandex/AliceAI-T5-35B-A0.6B).
• Optimised the production model, increasing throughput (RPS) by 5.6%.
• Reduced the runtime of the model optimisation pipeline before production deployment by approximately 40%.
• Researching encoder-decoder and decoder-only MoE LLM architectures.

Mentor

• Supporting students' academic and career development.

Senior AI Engineer, Corporate & Investment Banking

• Built two projects from scratch to pilots in four months — AI Copilot and autonomous agents for online trading.

Senior NLP Data Scientist

• Independently developed and deployed review summarisation (the “People Say” feature), increasing click-through conversion.
• Worked on developing generative summarisation for online search.
• Developed product-to-tag matching with macro F1 = 0.97 on a human-annotated dataset.
• Conducted technical interviews across NLP and adjacent ML teams, assessing candidates in NLP, ML engineering and applied data science.

ML Engineer, NLP B2C Team

• Developed the RLHF training stage for the GigaChat LLM and ran research experiments to improve model behaviour and alignment.
• Improved the SFT training pipeline for GigaChat and adapted internal models to product-specific requirements.
• Built pilot integrations of GigaChat into B2C and internal banking products.
• Developed retrieval components for RAG-based business use cases powered by GigaChat.

Data Scientist Intern, NLP B2C Team

• Worked on speech analytics for telephone calls, including clustering and visualisation of call data.
• Improved the intent recognition model training pipeline and adapted it to internal quality requirements.

PhD Candidate, Computer Science and Informatics

MSc, Computer Science; Machine Learning and Data Analysis

BSc, Applied Mathematics; Data Analysis