Description
The RnD NLP team is engaged in cutting-edge research in LLM development; areas of interest include improving LLM training methodologies (pretrain, SFT, RL), as well as exploring new technological components (reasoning, agents, specialized models, etc.)
We aim to bring research results to practical application and make the flagship GigaChat model smarter, better, and more useful.
Responsibilities
- working with large language models, deconstructing existing pipelines piece by piece, identifying areas where a diff of +1-10% can be made, generating research hypotheses where a diff of x2-x10 can be achieved
- training, validation, and implementation of new state-of-the-art approaches for relevant tasks
- packaging research and presenting results, delivery to GigaChat production
- publishing research results in top scientific conferences (optional).
Requirements
- experience in modern transformer-based NLP, excellent theoretical understanding of the development stages of top-tier industrial LLMs, practical experience (including pet projects) in at least one relevant stage
- good exposure to scientific and industrial articles, understanding of the industry's direction of development, which research areas are the "hottest," which methods/models/datasets achieve specific results, which benchmarks validate them, etc.
Will be a plus:
- extensive experience in classical ML/NLP
- A/A* level publications and experience writing articles
- prize-winning positions in machine learning competitions (Kaggle, etc.).
Conditions
- comfortable, modern office near Kutuzovskaya metro station
- opportunity to choose a convenient schedule – office/hybrid/remote (within Russia)
- annual bonus
- corporate gym and relaxation areas
- more than 400 educational programs from SberUniversity for professional and career development
- extended voluntary health insurance, preferential insurance for family, and corporate pension program
- employee mortgage under a discount program
- free SberPrime+ subscription, discounts on products from partner companies
- referral bonus for recommending friends to the Sber team