Reach out directly about this role
ML Researcher of Generative Personalization Models
We are solving the personalization task using LLM-like technologies: we map users into a specialized token space and process the resulting sequences using generative models. The results of our team's work are applied in all major Yandex recommendation systems.
Research in our team covers the following areas:
We also publish works at international conferences:
Our dream is to build a single-stage generative recommendation system and integrate it into an LLM as a new modality.
Optimize neural network architecture We are constantly evolving the architecture and training recipes of our models, adapting the latest industrial trends. Our task is not only to ensure high prediction quality but also to handle the load created by Yandex's many millions of users.
Develop tokenization algorithms Tokenization in generative personalization models is an actively developing area. It includes building semantic models, researching quantization algorithms, and continuous learning.
Research Alignment using RL We are building algorithms that adapt to user tastes using GRPO, learning from both reward models and direct feedback.
Integrate into LLM as a new modality We research not only generative personalization models but also the possibilities of combining them with LLMs to build systems with unique properties: the ability to follow instructions and explain recommendations.
More about ML at Yandex — in the channel Yandex for ML
3 years
Experience
Full-time
Employment
Data Science & ML
Specialization
IT & Tech
Industry
Corporation
Company Type
Corporation
Company Type